{ "S1sh2": { "input_fingerprint": "55b39eb03ce4af59", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 흑백 화면 속 머리에 피를 흘린 채 상체가 땅을 향해 반쯤 기울어진 자세로 무너지는 찰나의 조선인 여자.\n\nLOCATION (lock): An open patch within a winter reed field, among tall dry reeds in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fatally wounded Joseon woman's body collapses headfirst toward the ground in the reeds, with her upper body pitched steeply downward and her arms and legs hanging limp behind the falling torso; once down, her body must remain at that spot unless the scene explicitly relocates it.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 8, "A": 6, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "지문에서 요구한 '몸통 뒤로 축 늘어진 팔다리'와 '머리부터 고꾸라지며 쓰러지는 찰나'의 자세를 가장 정확하고 역동적으로 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "상체가 기울어진 모습은 표현되었으나, 다리가 몸 뒤로 늘어지지 않고 이미 바닥에 안정적으로 닿아 있어 '쓰러지는 찰나'의 느낌이 부족합니다." }, { "label": "B", "score": 4, "verdict_ko": "쓰러지는 찰나가 아닌 땅에 무릎을 꿇고 주저앉은 정적인 자세로 묘사되어 요구된 역동적 동작 지침을 충족하지 못했습니다." } ], "refs": [], "critique": { "issues": [ { "issue_ko": "여자가 오른팔로 바닥을 짚고 있어, 팔과 다리가 무너지는 몸통 뒤로 축 늘어져 있어야 한다는 지시 사항을 위반함.", "fix_en": "Reposition the right arm to hang limply backwards alongside the body instead of bracing against the ground." } ] }, "composition_critique": { "issues": [ { "fix_en": "Repose the woman in place into a steep headfirst fall toward the ground, with her upper torso pitched downward and both arms and legs hanging limp behind rather than supporting or catching her body.", "issue_ko": "여성이 머리부터 땅으로 급격히 무너지며 팔다리가 뒤로 축 늘어진 구도가 아니라, 팔과 무릎으로 몸을 지탱하며 기어가는 듯한 자세로 보인다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Reposition the right arm to hang limply backwards alongside the body instead of bracing against the ground.\n- Repose the woman in place into a steep headfirst fall toward the ground, with her upper torso pitched downward and both arms and legs hanging limp behind rather than supporting or catching her body.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "흑백 화면 속 쓰러지는 조선 여인의 모습은 잘 연출되었으나, 팔이 몸통 뒤로 축 늘어져야 한다는 지시와 달리 왼팔로 땅을 짚고 있습니다." }, { "label": "B", "score": 3, "verdict_ko": "팔이 늘어져야 한다는 지시를 어기고 땅을 짚고 있으며, 해당 왼손의 구조가 심하게 일그러진 해부학적 오류가 발생했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "흑백 화면 속 쓰러지는 조선 여인의 모습은 잘 연출되었으나, 팔이 몸통 뒤로 축 늘어져야 한다는 지시와 달리 왼팔로 땅을 짚고 있습니다." }, { "label": "B", "score": 3, "verdict_ko": "팔이 늘어져야 한다는 지시를 어기고 땅을 짚고 있으며, 해당 왼손의 구조가 심하게 일그러진 해부학적 오류가 발생했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "오른팔이 바닥을 짚고 있어 '팔이 뒤로 축 늘어진' 영문 지시를 완전히 따르진 않았으나, B에 비해 신체 해부학적 구조가 정상적입니다." }, { "label": "B", "score": 4, "verdict_ko": "마찬가지로 오른팔 지시를 위반했으며, 화면 우측에 보이는 왼손과 팔뚝 부위의 해부학적 구조가 기형적으로 생성되는 치명적인 오류가 있습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "오른팔이 바닥을 짚고 있어 '팔이 뒤로 축 늘어진' 영문 지시를 완전히 따르진 않았으나, B에 비해 신체 해부학적 구조가 정상적입니다." }, { "label": "A", "score": 4, "verdict_ko": "마찬가지로 오른팔 지시를 위반했으며, 화면 우측에 보이는 왼손과 팔뚝 부위의 해부학적 구조가 기형적으로 생성되는 치명적인 오류가 있습니다." } ] }, "combined": { "totals": { "A": 10, "B": 9 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S1sh5": { "input_fingerprint": "e3b3d356a108f2f9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 갈대숲 너머에서 도망치는 조선인들을 향해 소총을 일제히 겨누고 사격 자세를 취한 30여 명의 일본인 군인 무리.\n\nLOCATION (lock): A separate stretch of the winter reed field behind the fleeing group, with clear firing lines through the reeds. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 4, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 '일제강점기' 시대 배경에 맞는 제국주의 군복과 소총을 정확히 구현했으며, 약 30명의 군인이 사격하는 모습을 지시문 그대로 충실히 연출했습니다." }, { "label": "C", "score": 4, "verdict_ko": "사격 대형과 배경 연출은 양호하나, 일제강점기가 아닌 임진왜란 시기의 갑옷과 조총을 그려 시대 배경을 오인했고 인원수도 부족합니다." }, { "label": "A", "score": 3, "verdict_ko": "시대적 배경을 잘못 구현했을 뿐만 아니라, 샷 텍스트에 없는 쓰러진 시신들을 화면 전경에 임의로 추가하여 '지시되지 않은 인물 등장 금지' 규칙을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S1sh2_sel.png" } ], "critique": { "issues": [ { "issue_ko": "왼쪽에서 두 번째 서 있는 군인의 소총 총열 부분에 허공에 뜬 손과 다른 소총의 파편이 비정상적으로 겹쳐 있습니다.", "fix_en": "Remove the floating hand and extra rifle fragment intersecting the barrel of the second standing soldier from the left.", "unfixable": false }, { "issue_ko": "맨 앞줄 오른쪽 끝에서 무릎을 꿇고 있는 군인의 소총 앞부분(총열)이 중간에 끊기듯 사라져 있습니다.", "fix_en": "Extend the rifle barrel of the rightmost kneeling soldier in the front row so it is whole and complete.", "unfixable": false }, { "issue_ko": "왼쪽에서 세 번째 서 있는 군인이 소총을 쥐고 있는 왼손이 손가락이 전혀 묘사되지 않은 뭉개진 덩어리 형태입니다.", "fix_en": "Render distinct, anatomically correct fingers on the left hand of the third standing soldier from the left.", "unfixable": false }, { "issue_ko": "뒷줄 배경에 서 있는 다수의 군인들 얼굴이 눈, 코, 입 등의 이목구비가 없는 형체로 뭉개져 있어 '모든 인물을 온전한 사람으로 묘사하라'는 지시를 위반했습니다.", "fix_en": "Restore distinct and anatomically correct facial features to the soldiers standing in the background rows.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "issue_ko": "후열의 여러 군인이 소총을 겨누지 않고 선 채로 있어, 30여 명 전원이 도망자들을 향해 일제히 사격 자세를 취한다는 구성과 어긋난다.", "fix_en": "Raise the rear soldiers’ rifles into firing positions aimed along the same screen-right axis as the front ranks, preserving every figure’s current position and the existing framing.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the floating hand and extra rifle fragment intersecting the barrel of the second standing soldier from the left.\n- Extend the rifle barrel of the rightmost kneeling soldier in the front row so it is whole and complete.\n- Render distinct, anatomically correct fingers on the left hand of the third standing soldier from the left.\n- Restore distinct and anatomically correct facial features to the soldiers standing in the background rows.\n- Raise the rear soldiers’ rifles into firing positions aimed along the same screen-right axis as the front ranks, preserving every figure’s current position and the existing framing.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "프롬프트에 명시된 갈대숲 배경, 30여 명의 일본군, 일제히 겨눈 사격 자세를 매우 사실적으로 잘 구현했습니다." }, { "label": "B", "score": 8, "verdict_ko": "후보 A와 동일한 이미지로, 갈대숲과 사격 자세를 취한 다수의 일본군을 프롬프트에 맞게 훌륭하게 표현했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "프롬프트에 명시된 갈대숲 배경, 30여 명의 일본군, 일제히 겨눈 사격 자세를 매우 사실적으로 잘 구현했습니다." }, { "label": "B", "score": 8, "verdict_ko": "후보 A와 동일한 이미지로, 갈대숲과 사격 자세를 취한 다수의 일본군을 프롬프트에 맞게 훌륭하게 표현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "이전 샷의 흑백 톤과 겨울 갈대밭 배경을 완벽히 유지하며, 지시문대로 30여 명의 군인들이 일제히 사격 자세를 취하는 모습을 사실적으로 연출했습니다." }, { "label": "B", "score": 7, "verdict_ko": "A와 시각적으로 동일한 이미지로, 이전 샷의 환경적 맥락과 질감을 잘 이어받아 일본군 무리의 조준 자세를 정확하게 묘사했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "이전 샷의 흑백 톤과 겨울 갈대밭 배경을 완벽히 유지하며, 지시문대로 30여 명의 군인들이 일제히 사격 자세를 취하는 모습을 사실적으로 연출했습니다." }, { "label": "A", "score": 7, "verdict_ko": "A와 시각적으로 동일한 이미지로, 이전 샷의 환경적 맥락과 질감을 잘 이어받아 일본군 무리의 조준 자세를 정확하게 묘사했습니다." } ] }, "combined": { "totals": { "A": 15, "B": 15 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S1sh2" } }, "S2sh1::bgfirst_bg": { "input_fingerprint": "2d551576d8800d26", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 판자가 얼기설기 덧대어진 어두운 창고 안, 바닥에 웅크려 모인 채 양손으로 입을 꽉 틀어막고 눈물을 흘리는 조선인들의 모습.\n\nLOCATION (lock): Inside the cramped wooden storehouse, enclosed by roughly patched boards and dim in daytime.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From near standing eye level at the rear of the warehouse, a static wide composition places the crouching Koreans in a compressed cluster across the lower middle of frame. The rough interior structure encloses them on all sides, while the door lies beyond the group on the same axis that the following dolly will take.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: crouching Korean survivors in the lower-center of the frame, midground; warehouse door in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 덧댄 판자와 각목 벽 (Roughly pieced together as described); used as The visible gaps and overlapping boards compress the group within an improvised enclosure; 창고 문 (Closed before the soldiers enter) — The interior-facing side is visible beyond the group; used as Acts as the distant threat axis beyond the crouching survivors.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 판자가 얼기설기 덧대어진 어두운 창고 안, 바닥에 웅크려 모인 채 양손으로 입을 꽉 틀어막고 눈물을 흘리는 조선인들의 모습.\n\nLOCATION (lock): Inside the cramped wooden storehouse, enclosed by roughly patched boards and dim in daytime.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From near standing eye level at the rear of the warehouse, a static wide composition places the crouching Koreans in a compressed cluster across the lower middle of frame. The rough interior structure encloses them on all sides, while the door lies beyond the group on the same axis that the following dolly will take.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: crouching Korean survivors in the lower-center of the frame, midground; warehouse door in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 덧댄 판자와 각목 벽 (Roughly pieced together as described); used as The visible gaps and overlapping boards compress the group within an improvised enclosure; 창고 문 (Closed before the soldiers enter) — The interior-facing side is visible beyond the group; used as Acts as the distant threat axis beyond the crouching survivors.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S2sh1__bgfirst_bg.png", "asset_id": "62b9d0d2-b934-4ecf-993b-5a1b5b31994e", "input_asset_ids": [ "ee393b5d-a7f2-4384-bf69-58f31d208031", "f66acb83-fc67-490a-b343-234a9f9db0c4" ] }, "S2sh1": { "input_fingerprint": "5540090d25fbb424", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 판자가 얼기설기 덧대어진 어두운 창고 안, 바닥에 웅크려 모인 채 양손으로 입을 꽉 틀어막고 눈물을 흘리는 조선인들의 모습.\n\nLOCATION (lock): Inside the cramped wooden storehouse, enclosed by roughly patched boards and dim in daytime. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 판자가 얼기설기 덧대어진 어두운 창고 안, 바닥에 웅크려 모인 채 양손으로 입을 꽉 틀어막고 눈물을 흘리는 조선인들의 모습.\n\nLOCATION (lock): Inside the cramped wooden storehouse, enclosed by roughly patched boards and dim in daytime. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 판자가 얼기설기 덧대어진 어두운 창고 안, 바닥에 웅크려 모인 채 양손으로 입을 꽉 틀어막고 눈물을 흘리는 조선인들의 모습.\n\nLOCATION (lock): Inside the cramped wooden storehouse, enclosed by roughly patched boards and dim in daytime. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S2sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S2sh1.png" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L02B02.png" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "모든 인물이 '양손으로 입을 꽉 틀어막고 눈물을 흘리는' 지시를 충실히 구현했으며 공간의 분위기도 잘 살렸습니다." }, { "label": "A", "score": 4, "verdict_ko": "인물들이 대부분 한 손으로만 입을 가리고 있어 '양손으로'라는 핵심 행동 지시를 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "모든 인물이 '양손으로 입을 꽉 틀어막고 눈물을 흘리는' 지시를 충실히 구현했으며 공간의 분위기도 잘 살렸습니다." }, { "label": "A", "score": 4, "verdict_ko": "인물들이 대부분 한 손으로만 입을 가리고 있어 '양손으로'라는 핵심 행동 지시를 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "모든 인물이 '양손으로' 입을 틀어막고 눈물을 흘리는 지문을 충실히 구현했으며, 배경 레퍼런스도 정확히 반영했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물들이 웅크린 자세는 좋으나, 대부분 한 손으로만 입을 가리고 있어 '양손으로'라는 구체적인 행동 지시를 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "모든 인물이 '양손으로' 입을 틀어막고 눈물을 흘리는 지문을 충실히 구현했으며, 배경 레퍼런스도 정확히 반영했습니다." }, { "label": "A", "score": 4, "verdict_ko": "인물들이 웅크린 자세는 좋으나, 대부분 한 손으로만 입을 가리고 있어 '양손으로'라는 구체적인 행동 지시를 위반했습니다." } ] }, "combined": { "totals": { "A": 8, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 8, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "모든 인물이 '양손으로 입을 꽉 틀어막고 눈물을 흘리는' 지시를 충실히 구현했으며 공간의 분위기도 잘 살렸습니다." }, { "label": "A", "score": 4, "verdict_ko": "인물들이 대부분 한 손으로만 입을 가리고 있어 '양손으로'라는 핵심 행동 지시를 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L02B02.png" } ], "critique": { "issues": [ { "issue_ko": "왼쪽에서 두 번째에 있는 남성이 한 손으로만 입을 가리고 있어, '양손으로' 입을 꽉 틀어막으라는 지시를 위반했습니다.", "fix_en": "Raise the left hand of the man second from the left so that both of his hands are tightly covering his mouth.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Raise the left hand of the man second from the left so that both of his hands are tightly covering his mouth.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "A와 동일하게 좌측 남성(무릎 위의 세 번째 손)과 중앙 여성(팔을 잡은 여분의 손)에게 치명적인 신체 구조 오류가 있으나, 입을 막고 있는 손들의 형태는 상대적으로 자연스럽습니다." }, { "label": "A", "score": 3, "verdict_ko": "좌측 남성과 중앙 여성에게 여분의 손이 나타나는 치명적인 구조 오류가 있으며, 우측 여성의 손가락 개수와 형태가 B에 비해 심하게 일그러져 있습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "A와 동일하게 좌측 남성(무릎 위의 세 번째 손)과 중앙 여성(팔을 잡은 여분의 손)에게 치명적인 신체 구조 오류가 있으나, 입을 막고 있는 손들의 형태는 상대적으로 자연스럽습니다." }, { "label": "A", "score": 3, "verdict_ko": "좌측 남성과 중앙 여성에게 여분의 손이 나타나는 치명적인 구조 오류가 있으며, 우측 여성의 손가락 개수와 형태가 B에 비해 심하게 일그러져 있습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "어두운 창고라는 장소와 양손으로 입을 막고 우는 인물들의 자세, 시대적 복장이 프롬프트와 참조 이미지에 완벽히 부합합니다." }, { "label": "B", "score": 8, "verdict_ko": "후보 A와 동일한 이미지로, 장소와 인물의 감정 표현, 자세 및 소품 등 모든 지시사항을 훌륭하게 반영했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "어두운 창고라는 장소와 양손으로 입을 막고 우는 인물들의 자세, 시대적 복장이 프롬프트와 참조 이미지에 완벽히 부합합니다." }, { "label": "A", "score": 8, "verdict_ko": "후보 A와 동일한 이미지로, 장소와 인물의 감정 표현, 자세 및 소품 등 모든 지시사항을 훌륭하게 반영했습니다." } ] }, "combined": { "totals": { "A": 11, "B": 12 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S2sh1__bgfirst_bg.png", "bg_asset_id": "62b9d0d2-b934-4ecf-993b-5a1b5b31994e", "bg_record_key": "S2sh1::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S2sh5": { "input_fingerprint": "55354d484776763a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 허공으로 붕 뜬 채 보이지 않는 강한 힘에 의해 문 밖으로 거칠게 끌려나가는 일본인 장교의 몸.\n\nLOCATION (lock): The interior threshold of the wooden storehouse, immediately beside its open door to the reed field. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rough plank walls, dim daylight, and cramped makeshift interior. Exclude the huddled civilians from the doorway action and add only the officer being dragged outside.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 일본인 장교 (일본인 성인 남성, 날카로운 얼굴, 짧게 정돈한 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 4, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 실내 배경에서 보이지 않는 힘에 의해 공중으로 뜬 채 끌려나가는 장교의 모습을 지시대로 완벽하게 구현했습니다." }, { "label": "C", "score": 4, "verdict_ko": "보이지 않는 힘에 이끌려가는 동작은 잘 표현되었으나, 지시된 실내 공간이 아닌 외부 구도에서 촬영되어 감점되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "'보이지 않는 힘'이라는 명시적 지시와 달리 장교를 당기는 다른 인물의 팔이 프레임에 노출되어 치명적인 위반입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S2sh1_sel.png" } ], "critique": { "issues": [ { "issue_ko": "위로 치켜든 왼손의 손가락이 6개입니다.", "fix_en": "Remove the extra finger from the raised left hand so it has exactly five anatomically correct digits.", "unfixable": false }, { "issue_ko": "아래로 뻗은 오른손이 손가락 형태 없이 뭉개진 덩어리로 보입니다.", "fix_en": "Reconstruct the right hand to show clearly defined, anatomically correct fingers.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the extra finger from the raised left hand so it has exactly five anatomically correct digits.\n- Reconstruct the right hand to show clearly defined, anatomically correct fingers.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "보이지 않는 힘에 의해 밖으로 끌려가는 일본인 장교의 역동적인 공중 부양 묘사가 훌륭하며, 신체(특히 왼손)의 해부학적 디테일이 자연스럽습니다." }, { "label": "A", "score": 7, "verdict_ko": "지문과 로케이션의 요구사항은 잘 충족하였으나, 화면 좌측 하단으로 뻗은 장교의 왼손이 뭉개져 해부학적으로 부자연스럽게 묘사되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "보이지 않는 힘에 의해 밖으로 끌려가는 일본인 장교의 역동적인 공중 부양 묘사가 훌륭하며, 신체(특히 왼손)의 해부학적 디테일이 자연스럽습니다." }, { "label": "A", "score": 7, "verdict_ko": "지문과 로케이션의 요구사항은 잘 충족하였으나, 화면 좌측 하단으로 뻗은 장교의 왼손이 뭉개져 해부학적으로 부자연스럽게 묘사되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지시된 위치와 상황(보이지 않는 힘에 의해 끌려나가는 일본인 장교)을 매우 훌륭하게 연출했으며, 피사체의 신체 형태가 B에 비해 더 뚜렷하게 묘사되었습니다." }, { "label": "B", "score": 8, "verdict_ko": "지시된 상황과 인물 설정은 잘 반영되었으나, 아래로 뻗은 장교의 오른손이 뭉개져 덩어리(blob)처럼 표현된 점이 아쉽습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "지시된 위치와 상황(보이지 않는 힘에 의해 끌려나가는 일본인 장교)을 매우 훌륭하게 연출했으며, 피사체의 신체 형태가 B에 비해 더 뚜렷하게 묘사되었습니다." }, { "label": "A", "score": 8, "verdict_ko": "지시된 상황과 인물 설정은 잘 반영되었으나, 아래로 뻗은 장교의 오른손이 뭉개져 덩어리(blob)처럼 표현된 점이 아쉽습니다." } ] }, "combined": { "totals": { "A": 15, "B": 18 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S2sh1" } }, "S3sh5": { "input_fingerprint": "721139e1b911d80a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning suddenly to night, full moon.\n\nSHOT TEXT (authoritative, Korean): 가슴에서 피투성이 심장을 뽑아 든 채 직사각형 동공이 있는 황금빛 눈알을 번뜩이는 뿔 달린 그분(검은 연기 형체)의 얼굴.\n\nLOCATION (lock): On the bloodied ground immediately outside the reed-field storehouse. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The black entity holds the freshly torn-out heart, while the dying Japanese soldier remains down at its feet with his chest mutilated.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 6, "C": 4, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "제시문이 요구한 '얼굴' 프레이밍을 정확한 클로즈업으로 구현하여 1위이나, 검은 연기 형체와 황금빛 눈알 등 핵심 형태 특징을 누락한 점은 아쉽습니다." }, { "label": "C", "score": 4, "verdict_ko": "연기 형체, 황금빛 눈, 심장과 시체 등 설정은 가장 잘 묘사했으나, 프레이밍을 전신 롱샷으로 넓혀 얼굴에 집중하지 않은 점이 결정적 감점 요인입니다." }, { "label": "A", "score": 3, "verdict_ko": "프레이밍이 얼굴에 집중되지 않았으며, 연기 형체가 아닌 뿔 투구를 쓴 인간으로 잘못 묘사되어 우선순위 규칙에 따라 가장 낮은 평가를 받습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L03B01.png" } ], "critique": { "issues": [ { "issue_ko": "등장인물의 눈이 일반적인 둥근 동공과 어두운 색으로 묘사되어, 프롬프트에 명시된 '직사각형 동공이 있는 황금빛 눈알'이 반영되지 않았습니다.", "fix_en": "Change the irises to a bright golden color and reshape the pupils into rectangles.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Extend the frame slightly downward and wider to include the raised hand holding the heart while keeping the face dominant and preserving the existing camera axis.", "issue_ko": "프레이밍이 얼굴에 지나치게 밀착되어, 가슴에서 뽑아 들어 올린 심장과 이를 든 손이 화면 밖으로 빠져 핵심 동작이 보이지 않는다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the irises to a bright golden color and reshape the pupils into rectangles.\n- Extend the frame slightly downward and wider to include the raised hand holding the heart while keeping the face dominant and preserving the existing camera axis.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "뿔과 배경은 잘 표현되었으나, 텍스트에 명시된 황금빛 눈알, 직사각형 동공, 그리고 심장을 들고 있는 핵심적인 행동이 누락되었습니다." }, { "label": "B", "score": 8, "verdict_ko": "직사각형 동공을 가진 황금빛 눈알, 뿔, 그리고 프레임 하단에 심장을 들고 있는 모습까지 프롬프트의 요구사항을 충실히 반영했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "뿔과 배경은 잘 표현되었으나, 텍스트에 명시된 황금빛 눈알, 직사각형 동공, 그리고 심장을 들고 있는 핵심적인 행동이 누락되었습니다." }, { "label": "B", "score": 8, "verdict_ko": "직사각형 동공을 가진 황금빛 눈알, 뿔, 그리고 프레임 하단에 심장을 들고 있는 모습까지 프롬프트의 요구사항을 충실히 반영했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "직사각형 동공을 가진 황금빛 눈과 손에 든 피투성이 심장 등 텍스트의 핵심 지시사항을 정확하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "프롬프트가 명시한 황금빛 눈(직사각형 동공)과 손에 든 심장이 샷 내에 전혀 묘사되지 않아 핵심 요소를 누락했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "직사각형 동공을 가진 황금빛 눈과 손에 든 피투성이 심장 등 텍스트의 핵심 지시사항을 정확하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "프롬프트가 명시한 황금빛 눈(직사각형 동공)과 손에 든 심장이 샷 내에 전혀 묘사되지 않아 핵심 요소를 누락했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S3sh13": { "input_fingerprint": "8f5aa3309ea0416f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning suddenly to night, full moon.\n\nSHOT TEXT (authoritative, Korean): 앞으로 뻗어나온 그분(검은 연기 형체)의 손에 목이 강하게 틀어잡혀 얼굴 앞까지 바짝 당겨진 조선인 촌주의 놀란 표정.\n\nLOCATION (lock): The open ground before the reed-field storehouse, amid fallen soldiers. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The entity remains transfixed through the chest by the spear with the white antler-like tip, and the massacred Japanese soldiers remain scattered around the warehouse.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 조선인 촌주 (50대 중반 조선인 남성, 각진 중년 얼굴, 단정한 짙은 머리); 그분(응집된 검은 연기 형체) (검은 연기 인간형, 비정상적으로 긴 팔다리, 머리 위 두 뿔, 검은 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 8, "A": 5, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "'얼굴 앞까지 바짝 당겨진' 숏의 구도를 가장 정확히 연출했으며, 촌주의 머리띠와 창이 박힌 그분의 상태도 잘 유지했습니다." }, { "label": "A", "score": 5, "verdict_ko": "그분이 촌주를 얼굴 앞이 아닌 뒤에서 잡고 있어 연출 의도와 맞지 않으며, 촌주의 머리띠가 누락되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "지문에 명시되지 않은 제3의 뿔 달린 인물이 배경에 창을 들고 서 있어 심각한 캐릭터 추가 위반(Hard Violation)입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S3sh5_sel.png" }, { "label": "CHARACTER REFERENCE — 조선인 촌주: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 그분(응집된 검은 연기 형체): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "검은 연기 형체의 얼굴이 지정된 캐릭터 레퍼런스(균열이 간 악마 같은 외형, 빛나는 눈, 귀가 없음)와 달리 일반적인 사람의 이목구비와 귀를 가지고 있습니다.", "fix_en": "Replace the entity's human nose, lips, and ear with the faceless, cracked, demonic appearance and glowing orange eyes shown in its specific character reference.", "unfixable": false }, { "issue_ko": "괴물의 가슴을 관통한 창의 앞부분에 있는 하얀 사슴뿔 모양의 촉이 몸통 쪽을 향해 거꾸로 달려 있어 물리적으로 어색합니다.", "fix_en": "Reverse the orientation of the white antler tip on the spear so the prongs point outward and away from the entity's chest.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the entity's human nose, lips, and ear with the faceless, cracked, demonic appearance and glowing orange eyes shown in its specific character reference.\n- Reverse the orientation of the white antler tip on the spear so the prongs point outward and away from the entity's chest.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "제시된 캐릭터 레퍼런스에 맞춰 '그분'의 갈라진 얼굴과 빛나는 눈을 정확히 묘사했으며, 촌주가 멱살을 잡힌 채 놀라는 표정과 가슴에 꽂힌 창의 디테일도 훌륭합니다." }, { "label": "A", "score": 6, "verdict_ko": "상황과 연출은 프롬프트에 부합하나, '그분'의 얼굴이 캐릭터 레퍼런스의 비인간적인 형태를 따르지 않고 인간의 이목구비에 그을음이 묻은 형태로 잘못 렌더링되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "제시된 캐릭터 레퍼런스에 맞춰 '그분'의 갈라진 얼굴과 빛나는 눈을 정확히 묘사했으며, 촌주가 멱살을 잡힌 채 놀라는 표정과 가슴에 꽂힌 창의 디테일도 훌륭합니다." }, { "label": "A", "score": 6, "verdict_ko": "상황과 연출은 프롬프트에 부합하나, '그분'의 얼굴이 캐릭터 레퍼런스의 비인간적인 형태를 따르지 않고 인간의 이목구비에 그을음이 묻은 형태로 잘못 렌더링되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "'검은 연기 형체'의 레퍼런스 외형(갈라진 얼굴, 빛나는 눈)과 촌주의 표정을 지시된 샷 구도에 맞게 정확히 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "목을 조르는 구도는 적절하나, '검은 연기 형체'의 얼굴이 레퍼런스와 달리 인간의 이목구비로 표현되어 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "'검은 연기 형체'의 레퍼런스 외형(갈라진 얼굴, 빛나는 눈)과 촌주의 표정을 지시된 샷 구도에 맞게 정확히 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "목을 조르는 구도는 적절하나, '검은 연기 형체'의 얼굴이 레퍼런스와 달리 인간의 이목구비로 표현되어 감점되었습니다." } ] }, "combined": { "totals": { "A": 12, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S3sh5" } }, "S3sh17": { "input_fingerprint": "ede366b8fd3658c7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning suddenly to night, full moon.\n\nSHOT TEXT (authoritative, Korean): 창고에서 나온 조선인들을 향해 고개를 돌린 조선인 촌주의 두 눈이 짐승처럼 황금빛으로 번뜩이는 얼굴.\n\nLOCATION (lock): Just outside the storehouse entrance, where the survivors emerge into the reed field. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The black entity and the spear embedded in it have vanished, while the Japanese soldiers' bloodied bodies remain scattered outside the warehouse.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 조선인 촌주 (각성 상태) (50대 중반 조선인 남성, 각진 중년 얼굴, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "C": 5, "B": 3 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "이전 샷의 배경, 시체, 머리띠를 정확히 유지하고 훌륭한 구도를 보여주나, 비유적 표현을 물리적인 안광으로 묘사한 점은 지시 위반입니다." }, { "label": "C", "score": 5, "verdict_ko": "배경과 시체 등은 유지되었으나, 안광 묘사 규정을 위반했고 촌주가 바라보는 방향과 배경 인물들의 위치가 논리적으로 어긋납니다." }, { "label": "B", "score": 3, "verdict_ko": "눈의 사실적 묘사는 규정을 준수했으나, 이전 샷의 시체와 머리띠가 누락되고 원본에 없는 수레가 등장하여 치명적인 위반입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S3sh13_sel.png" }, { "label": "CHARACTER REFERENCE — 조선인 촌주 (각성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물의 두 눈이 문자 그대로 황금빛으로 발광하고 있습니다. 이는 비유적 표현을 연기로 구현하라는 지침과, 인간의 안구를 해부학적으로 변형하거나 스스로 빛나게 해서는 안 된다는 지침을 위반했습니다.", "fix_en": "Replace the glowing yellow effect in the eyes with normal, anatomically correct human eyes featuring natural white sclera and natural dark irises; convey the 'flashing' effect purely through external lighting reflecting on normal wet eyes.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the glowing yellow effect in the eyes with normal, anatomically correct human eyes featuring natural white sclera and natural dark irises; convey the 'flashing' effect purely through external lighting reflecting on normal wet eyes.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "눈이 황금빛으로 번뜩인다는 비유적 표현을 물리적인 안광 대신 조명과 연기로 사실적으로 잘 구현했으며 시신들의 배치도 정확합니다." }, { "label": "A", "score": 3, "verdict_ko": "인물의 눈이 문자 그대로 빛나는 판타지적 CG 효과로 묘사되어, 살아있는 인간의 신체를 변형하지 말라는 핵심 지침을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "눈이 황금빛으로 번뜩인다는 비유적 표현을 물리적인 안광 대신 조명과 연기로 사실적으로 잘 구현했으며 시신들의 배치도 정확합니다." }, { "label": "A", "score": 3, "verdict_ko": "인물의 눈이 문자 그대로 빛나는 판타지적 CG 효과로 묘사되어, 살아있는 인간의 신체를 변형하지 말라는 핵심 지침을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "은유적 표현(황금빛으로 번뜩이는 눈)을 해부학적 변형 없이 사실적인 조명과 연기로 훌륭하게 표현했으며, 이전 샷의 상태를 잘 유지했습니다." }, { "label": "B", "score": 3, "verdict_ko": "인물의 눈이 문자 그대로 빛나는 판타지 그래픽으로 묘사되어, 은유적 표현을 사실적으로 연출하라는 지침을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "은유적 표현(황금빛으로 번뜩이는 눈)을 해부학적 변형 없이 사실적인 조명과 연기로 훌륭하게 표현했으며, 이전 샷의 상태를 잘 유지했습니다." }, { "label": "A", "score": 3, "verdict_ko": "인물의 눈이 문자 그대로 빛나는 판타지 그래픽으로 묘사되어, 은유적 표현을 사실적으로 연출하라는 지침을 위반했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S3sh13" } }, "groupbg::퇴근_버스_객실": { "input_fingerprint": "bcf4396f22f11d49", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "퇴근_버스_객실", "tags": [ "S4sh3", "S4sh8" ] }, "context_sig": "f0e2606a12a065cd" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A window-side passenger seat inside a moving city commuter bus at night, lit by the phone and passing urban lights.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n서울 시내 퇴근 버스 내부: 현대 도심을 달리는 버스의 빽빽한 실내. 차창 밖으로 네온사인 불빛이 스친다. (특징: 유리창에 비치는 도심 네온사인 불빛; 좌석에 앉거나 서 있는 사람들의 밀집된 모습; 스마트폰 화면 불빛들)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실내. 달리는 퇴근 버스 - 밤\n- 차창 밖으로 도심 네온싸인, 잡다한 불빛들이 어지러이 스쳐 지나간다.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A window-side passenger seat inside a moving city commuter bus at night, lit by the phone and passing urban lights.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n서울 시내 퇴근 버스 내부: 현대 도심을 달리는 버스의 빽빽한 실내. 차창 밖으로 네온사인 불빛이 스친다. (특징: 유리창에 비치는 도심 네온사인 불빛; 좌석에 앉거나 서 있는 사람들의 밀집된 모습; 스마트폰 화면 불빛들)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실내. 달리는 퇴근 버스 - 밤\n- 차창 밖으로 도심 네온싸인, 잡다한 불빛들이 어지러이 스쳐 지나간다.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_퇴근_버스_객실_26a6f7.png", "asset_id": "8bc7ab4b-18d8-45ee-bb7b-b8c91222b03b", "input_asset_ids": [ "1c271367-434e-4b68-8422-1af97a53f483" ], "origin_tag": "S4sh3", "place_text": "A window-side passenger seat inside a moving city commuter bus at night, lit by the phone and passing urban lights.", "origin_inputs": { "place_text": "A window-side passenger seat inside a moving city commuter bus at night, lit by the phone and passing urban lights.", "time_of_day_en": "night", "conti_asset_id": "1c271367-434e-4b68-8422-1af97a53f483" } }, "S4sh3::bgfirst_bg": { "input_fingerprint": "75f652d90e6a7f8c", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 수십 개의 부재중 전화 기록이 빼곡히 채워진 핸드폰 화면을 보며 미간을 찌푸린 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): A window-side passenger seat inside a moving city commuter bus at night, lit by the phone and passing urban lights.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the aisle edge slightly above seated eye level, the camera pushes into an oblique close composition that places 김지후(강수리영)’s furrowed three-quarter face in the upper half and the tilted phone screen in the lower third. Sufficient depth keeps the missed-call list readable while her hand and the seat edge provide scale and prevent the device from becoming an isolated insert.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: active phone screen in the lower-center of the frame, foreground; 김지후(강수리영) in the upper-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 핸드폰 화면 (On, displaying numerous unanswered calls from her mother) — The active front display faces the camera at a readable angle and shows the densely filled missed-call history; used as Shares focus with 김지후(강수리영)’s expression and carries the missed-call narrative information; 버스 좌석 (Occupied by 김지후(강수리영)) — The aisle-facing edge and backrest frame her seated posture; used as Creates a narrow foreground boundary along the aisle side of her body; 차창 (Beside her window seat) — The window shows passing urban neon signs and miscellaneous city lights outside; used as Supplies moving night context beyond her row without displacing the phone as the focus.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 수십 개의 부재중 전화 기록이 빼곡히 채워진 핸드폰 화면을 보며 미간을 찌푸린 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): A window-side passenger seat inside a moving city commuter bus at night, lit by the phone and passing urban lights.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the aisle edge slightly above seated eye level, the camera pushes into an oblique close composition that places 김지후(강수리영)’s furrowed three-quarter face in the upper half and the tilted phone screen in the lower third. Sufficient depth keeps the missed-call list readable while her hand and the seat edge provide scale and prevent the device from becoming an isolated insert.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: active phone screen in the lower-center of the frame, foreground; 김지후(강수리영) in the upper-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 핸드폰 화면 (On, displaying numerous unanswered calls from her mother) — The active front display faces the camera at a readable angle and shows the densely filled missed-call history; used as Shares focus with 김지후(강수리영)’s expression and carries the missed-call narrative information; 버스 좌석 (Occupied by 김지후(강수리영)) — The aisle-facing edge and backrest frame her seated posture; used as Creates a narrow foreground boundary along the aisle side of her body; 차창 (Beside her window seat) — The window shows passing urban neon signs and miscellaneous city lights outside; used as Supplies moving night context beyond her row without displacing the phone as the focus.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S4sh3__bgfirst_bg.png", "asset_id": "44d43f07-32d2-4a48-ac54-fb7673d99440", "input_asset_ids": [ "1c271367-434e-4b68-8422-1af97a53f483", "8bc7ab4b-18d8-45ee-bb7b-b8c91222b03b" ] }, "S4sh3": { "input_fingerprint": "6538fe0438db034d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수십 개의 부재중 전화 기록이 빼곡히 채워진 핸드폰 화면을 보며 미간을 찌푸린 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): A window-side passenger seat inside a moving city commuter bus at night, lit by the phone and passing urban lights. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear the arrowhead-shaped animal-bone necklace that she has worn since her first appearance.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수십 개의 부재중 전화 기록이 빼곡히 채워진 핸드폰 화면을 보며 미간을 찌푸린 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): A window-side passenger seat inside a moving city commuter bus at night, lit by the phone and passing urban lights. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear the arrowhead-shaped animal-bone necklace that she has worn since her first appearance.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수십 개의 부재중 전화 기록이 빼곡히 채워진 핸드폰 화면을 보며 미간을 찌푸린 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): A window-side passenger seat inside a moving city commuter bus at night, lit by the phone and passing urban lights. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear the arrowhead-shaped animal-bone necklace that she has worn since her first appearance.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S4sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S4sh3.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_퇴근_버스_객실_26a6f7.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "인물의 찌푸린 표정과 기기 조작 방향이 지문 및 물리적 현실에 부합하며, 지문에 없는 추가 인물 없이 정확하게 연출되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "지문에 명시되지 않은 추가 승객들이 등장해 치명적인 감점 요소가 되었으며, 인물이 보는 스마트폰 화면이 카메라를 향하고 있어 조작 방향이 어색합니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "인물의 찌푸린 표정과 기기 조작 방향이 지문 및 물리적 현실에 부합하며, 지문에 없는 추가 인물 없이 정확하게 연출되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "지문에 명시되지 않은 추가 승객들이 등장해 치명적인 감점 요소가 되었으며, 인물이 보는 스마트폰 화면이 카메라를 향하고 있어 조작 방향이 어색합니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 인물, 복장, 배경을 정확히 구현했으며 스마트폰의 방향 규칙을 준수했으나 목걸이 형태가 다릅니다." }, { "label": "B", "score": 3, "verdict_ko": "지문에 없는 승객들이 배경에 등장하고 스마트폰 화면이 카메라를 향해 있어 심각한 규정 위반(하드 비올레이션)입니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 인물, 복장, 배경을 정확히 구현했으며 스마트폰의 방향 규칙을 준수했으나 목걸이 형태가 다릅니다." }, { "label": "A", "score": 3, "verdict_ko": "지문에 없는 승객들이 배경에 등장하고 스마트폰 화면이 카메라를 향해 있어 심각한 규정 위반(하드 비올레이션)입니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "인물의 찌푸린 표정과 기기 조작 방향이 지문 및 물리적 현실에 부합하며, 지문에 없는 추가 인물 없이 정확하게 연출되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "지문에 명시되지 않은 추가 승객들이 등장해 치명적인 감점 요소가 되었으며, 인물이 보는 스마트폰 화면이 카메라를 향하고 있어 조작 방향이 어색합니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_퇴근_버스_객실_26a6f7.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프롬프트에서 모든 텍스트 생성을 금지했으나, 캐릭터의 파란색 조끼와 창밖 네온사인에 문자가 생성되었습니다.", "fix_en": "Remove the white text from the blue vest and the glowing text from the neon signs outside the window.", "unfixable": false }, { "issue_ko": "목걸이 펜던트가 프롬프트에 명시된 '화살촉 모양(arrowhead-shaped)'이 아닌 가로형 뼈 장식입니다.", "fix_en": "Change the necklace pendant to a single, downward-pointing, arrowhead-shaped piece of bone.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the white text from the blue vest and the glowing text from the neon signs outside the window.\n- Change the necklace pendant to a single, downward-pointing, arrowhead-shaped piece of bone.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "인물의 표정과 배경은 훌륭하나, 필수 유지 항목인 '화살촉 모양의 뼈 목걸이' 대신 전혀 다른 형태의 목걸이가 묘사되어 지시도를 낮춥니다." }, { "label": "B", "score": 8, "verdict_ko": "지시된 구도와 찡그린 표정을 사실적으로 담아냈으며, 특히 화살촉 모양의 뼈 목걸이 설정을 정확하게 반영하여 가장 우수한 결과를 보여줍니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "인물의 표정과 배경은 훌륭하나, 필수 유지 항목인 '화살촉 모양의 뼈 목걸이' 대신 전혀 다른 형태의 목걸이가 묘사되어 지시도를 낮춥니다." }, { "label": "B", "score": 8, "verdict_ko": "지시된 구도와 찡그린 표정을 사실적으로 담아냈으며, 특히 화살촉 모양의 뼈 목걸이 설정을 정확하게 반영하여 가장 우수한 결과를 보여줍니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 화살촉 모양의 뼈 목걸이를 정확히 착용하고 있으며, 인물의 표정과 버스 배경 연출이 프롬프트에 부합합니다." }, { "label": "B", "score": 5, "verdict_ko": "표정과 배경 연출은 우수하나, 명시된 화살촉 모양이 아닌 다른 형태의 목걸이를 착용하여 디테일에서 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 화살촉 모양의 뼈 목걸이를 정확히 착용하고 있으며, 인물의 표정과 버스 배경 연출이 프롬프트에 부합합니다." }, { "label": "A", "score": 5, "verdict_ko": "표정과 배경 연출은 우수하나, 명시된 화살촉 모양이 아닌 다른 형태의 목걸이를 착용하여 디테일에서 감점되었습니다." } ] }, "combined": { "totals": { "A": 10, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S4sh3__bgfirst_bg.png", "bg_asset_id": "44d43f07-32d2-4a48-ac54-fb7673d99440", "bg_record_key": "S4sh3::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "퇴근_버스_객실", "groupbg_asset_id": "8bc7ab4b-18d8-45ee-bb7b-b8c91222b03b" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S4sh8": { "input_fingerprint": "6c84b6c411d12ab3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 차창에 맺힌 기괴한 형상을 전혀 눈치채지 못한 채 바닥만 내려다보는 김지후(강수리영)와 창문에 비친 그분(검은 연기 형체)의 황금색 눈알.\n\nLOCATION (lock): Inside the moving commuter bus beside a window reflecting passing city lights and a fleeting dark apparition. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her arrowhead-shaped animal-bone necklace; the smoky golden-eyed reflection exists only briefly on the bus window and leaves no lasting mark.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 그분(차창 환영 상태) (검은 연기 얼굴 형상, 차창에 맺힌 흐릿한 윤곽, 사람 같은 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 4, "C": 2 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "바닥을 내려다보는 시선 처리와 창문에 비친 황금색 눈의 연기 형상을 지시문대로 정확히 연출했습니다." }, { "label": "A", "score": 4, "verdict_ko": "바닥이 아닌 스마트폰을 보고 있어 지시문과 어긋나며, 이전 컷의 자세와 구도를 그대로 답습했습니다." }, { "label": "C", "score": 2, "verdict_ko": "창문 외에 버스 내부 허공에도 거대한 유령 얼굴이 중복으로 생성되어 치명적인 오류입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S4sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 그분(차창 환영 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "지후의 조끼 왼쪽 가슴에 있던 로고, 명찰, 펜들이 사라졌으며, 명찰이 조끼 하단으로 잘못 이동했습니다.", "fix_en": "Restore the logo, name badge, and pens to the left breast pocket of the blue vest, and remove the misplaced badge near the bottom right edge.", "unfixable": false }, { "issue_ko": "화살촉 목걸이 줄 외에 목을 꽉 조이는 불필요한 검은색 초커(여분의 줄)가 추가되었습니다.", "fix_en": "Remove the extra tight black line/cord around the neck, leaving only the single long cord attached to the arrowhead necklace.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the logo, name badge, and pens to the left breast pocket of the blue vest, and remove the misplaced badge near the bottom right edge.\n- Remove the extra tight black line/cord around the neck, leaving only the single long cord attached to the arrowhead necklace.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "바닥을 내려다보는 시선, 창문에 비친 연기 형상과 황금색 눈을 잘 구현했으며 이전 샷의 의상 상태(주머니에 펜과 명찰 없음)를 정확히 유지했습니다." }, { "label": "B", "score": 6, "verdict_ko": "요구된 주요 상황은 잘 표현했으나, 이전 샷에는 없던 명찰(정체불명의 텍스트 포함)과 펜들이 조끼에 추가되어 샷 간 연속성을 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "바닥을 내려다보는 시선, 창문에 비친 연기 형상과 황금색 눈을 잘 구현했으며 이전 샷의 의상 상태(주머니에 펜과 명찰 없음)를 정확히 유지했습니다." }, { "label": "B", "score": 6, "verdict_ko": "요구된 주요 상황은 잘 표현했으나, 이전 샷에는 없던 명찰(정체불명의 텍스트 포함)과 펜들이 조끼에 추가되어 샷 간 연속성을 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "이전 샷의 복장(명찰과 펜이 없는 조끼)과 연속성을 더 잘 유지하며, 환영의 황금색 눈과 시선 방향을 정확히 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "지시된 상황과 인물의 특징을 잘 반영했으나, 조끼에 이전 샷에 없던 불필요한 명찰과 펜이 추가되고 글씨가 깨져 보입니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "이전 샷의 복장(명찰과 펜이 없는 조끼)과 연속성을 더 잘 유지하며, 환영의 황금색 눈과 시선 방향을 정확히 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "지시된 상황과 인물의 특징을 잘 반영했으나, 조끼에 이전 샷에 없던 불필요한 명찰과 펜이 추가되고 글씨가 깨져 보입니다." } ] }, "combined": { "totals": { "A": 15, "B": 12 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S4sh3" } }, "S5sh1::bgfirst_bg": { "input_fingerprint": "e4c60249430d5cec", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 캔 음료가 가득한 편의점 냉장고 뒤편 좁은 공간에서 유통기한이 지난 도시락을 손에 꽉 쥔 김지후(강수리영)의 상체.\n\nLOCATION (lock): The narrow service space behind a convenience-store beverage refrigerator, screened from the sales floor by rows of cans.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height in the narrow space behind the stocked refrigerator rows, hold a tight medium view of 김지후(강수리영) with her upper body on the right half and the expired lunch box prominent near the lower center. Keep the camera static and slightly off her frontal axis, using the drink rows along the left edge to compress the space without obscuring her drained expression and blue vest.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: rows of canned drinks (packed tightly inside the refrigerator) — The cylindrical sides and visible fronts face obliquely toward the camera from the left edge; used as The tightly packed rows form a foreground edge that visually confines the character; expired lunch box (expired and held tightly) — Its upper and near side face the camera while it is clenched in both hands; used as Held near the lower center as the concrete evidence of her hurried late meal; basket of discarded food (filled with food marked for disposal) — Its open top angles toward the camera behind the lunch box; used as Provides contextual depth behind and below the primary figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 캔 음료가 가득한 편의점 냉장고 뒤편 좁은 공간에서 유통기한이 지난 도시락을 손에 꽉 쥔 김지후(강수리영)의 상체.\n\nLOCATION (lock): The narrow service space behind a convenience-store beverage refrigerator, screened from the sales floor by rows of cans.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height in the narrow space behind the stocked refrigerator rows, hold a tight medium view of 김지후(강수리영) with her upper body on the right half and the expired lunch box prominent near the lower center. Keep the camera static and slightly off her frontal axis, using the drink rows along the left edge to compress the space without obscuring her drained expression and blue vest.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: rows of canned drinks (packed tightly inside the refrigerator) — The cylindrical sides and visible fronts face obliquely toward the camera from the left edge; used as The tightly packed rows form a foreground edge that visually confines the character; expired lunch box (expired and held tightly) — Its upper and near side face the camera while it is clenched in both hands; used as Held near the lower center as the concrete evidence of her hurried late meal; basket of discarded food (filled with food marked for disposal) — Its open top angles toward the camera behind the lunch box; used as Provides contextual depth behind and below the primary figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S5sh1__bgfirst_bg.png", "asset_id": "cbf7aa21-2e1d-4f36-91f6-f0bfc80c777c", "input_asset_ids": [ "1067f982-dd47-4941-906a-08f535e2715c", "fcb32ffa-23ea-4815-9336-659e1c63479c" ] }, "S5sh1": { "input_fingerprint": "da8078566c6120a8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 캔 음료가 가득한 편의점 냉장고 뒤편 좁은 공간에서 유통기한이 지난 도시락을 손에 꽉 쥔 김지후(강수리영)의 상체.\n\nLOCATION (lock): The narrow service space behind a convenience-store beverage refrigerator, screened from the sales floor by rows of cans. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped animal-bone necklace while holding the expired lunchbox.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 캔 음료가 가득한 편의점 냉장고 뒤편 좁은 공간에서 유통기한이 지난 도시락을 손에 꽉 쥔 김지후(강수리영)의 상체.\n\nLOCATION (lock): The narrow service space behind a convenience-store beverage refrigerator, screened from the sales floor by rows of cans. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped animal-bone necklace while holding the expired lunchbox.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 캔 음료가 가득한 편의점 냉장고 뒤편 좁은 공간에서 유통기한이 지난 도시락을 손에 꽉 쥔 김지후(강수리영)의 상체.\n\nLOCATION (lock): The narrow service space behind a convenience-store beverage refrigerator, screened from the sales floor by rows of cans. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped animal-bone necklace while holding the expired lunchbox.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S5sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S5sh1.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L06B03.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 좁은 공간 내부에 카메라를 배치하여 샷의 배경을 정확히 구현했으며, 인물의 상체 프레이밍과 목걸이, 도시락을 쥔 자세가 자연스럽습니다." }, { "label": "B", "score": 4, "verdict_ko": "카메라가 지정된 공간 내부가 아닌 냉장고 선반 쪽에 위치해 시점이 어긋났으며, 도시락에 부자연스러운 영문 텍스트가 생성되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 좁은 공간 내부에 카메라를 배치하여 샷의 배경을 정확히 구현했으며, 인물의 상체 프레이밍과 목걸이, 도시락을 쥔 자세가 자연스럽습니다." }, { "label": "B", "score": 4, "verdict_ko": "카메라가 지정된 공간 내부가 아닌 냉장고 선반 쪽에 위치해 시점이 어긋났으며, 도시락에 부자연스러운 영문 텍스트가 생성되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 장소와 인물의 외양을 정확히 구현했으며, 인물이 소품(도시락)을 자연스럽게 자신의 시선 방향으로 들고 있어 소품 연출 지침을 훌륭히 준수했습니다." }, { "label": "A", "score": 5, "verdict_ko": "인물이 도시락의 라벨 면을 시선과 무관하게 카메라 쪽으로 보여주듯 들고 있어 소품 방향 지침을 위반했으며, 라벨에 프롬프트의 영문 번역(expired)이 변형된 텍스트로 유출되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 장소와 인물의 외양을 정확히 구현했으며, 인물이 소품(도시락)을 자연스럽게 자신의 시선 방향으로 들고 있어 소품 연출 지침을 훌륭히 준수했습니다." }, { "label": "B", "score": 5, "verdict_ko": "인물이 도시락의 라벨 면을 시선과 무관하게 카메라 쪽으로 보여주듯 들고 있어 소품 방향 지침을 위반했으며, 라벨에 프롬프트의 영문 번역(expired)이 변형된 텍스트로 유출되었습니다." } ] }, "combined": { "totals": { "A": 15, "B": 9 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 15, "B": 9 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 좁은 공간 내부에 카메라를 배치하여 샷의 배경을 정확히 구현했으며, 인물의 상체 프레이밍과 목걸이, 도시락을 쥔 자세가 자연스럽습니다." }, { "label": "B", "score": 4, "verdict_ko": "카메라가 지정된 공간 내부가 아닌 냉장고 선반 쪽에 위치해 시점이 어긋났으며, 도시락에 부자연스러운 영문 텍스트가 생성되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L06B03.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "캐릭터가 입고 있는 파란색 조끼에 레퍼런스 이미지에 있는 명찰, 주머니에 꽂힌 펜, 가슴 로고, 금속 단추 등의 디테일이 모두 누락되었으며 디자인이 다릅니다.", "fix_en": "Modify the blue vest to match the character reference exactly by adding the nametag, pens in the left pocket, chest logo, metal buttons, and pocket seams.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Match the reference framing as closely as possible with a global 2D crop and alignment while preserving the existing scene geometry.", "issue_ko": "배경의 카메라 프레이밍과 원근이 기준 사진과 일치하지 않아 왼쪽 기둥, 냉장고 열, 전경 바구니의 화면상 위치와 크기가 달라졌다.", "unfixable": true }, { "fix_en": "Scale Jihu up slightly and shift her composited figure to the right and downward, placing her head, torso, hands, and lunchbox at the layout-sketch coordinates.", "issue_ko": "지후와 도시락이 레이아웃 스케치보다 화면에서 작고 왼쪽 위에 배치되어 있으며, 특히 몸의 오른쪽 윤곽과 도시락이 목표 위치에 못 미친다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Modify the blue vest to match the character reference exactly by adding the nametag, pens in the left pocket, chest logo, metal buttons, and pocket seams.\n- Scale Jihu up slightly and shift her composited figure to the right and downward, placing her head, torso, hands, and lunchbox at the layout-sketch coordinates.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "캐릭터 레퍼런스의 조끼 디테일(명찰, 펜)을 더 충실하게 반영하였고, 프롬프트의 상황과 레이아웃을 정확히 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "전반적인 구도와 인상, 목걸이 등은 잘 일치하나 캐릭터의 조끼 디테일(주머니, 명찰, 펜)이 누락되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "캐릭터 레퍼런스의 조끼 디테일(명찰, 펜)을 더 충실하게 반영하였고, 프롬프트의 상황과 레이아웃을 정확히 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "전반적인 구도와 인상, 목걸이 등은 잘 일치하나 캐릭터의 조끼 디테일(주머니, 명찰, 펜)이 누락되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "배경과 레이아웃을 정확히 따랐으며 캐릭터 의상의 구조도 잘 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "배경과 구도는 일치하나 캐릭터 조끼의 구조적 디테일이 심하게 누락되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "배경과 레이아웃을 정확히 따랐으며 캐릭터 의상의 구조도 잘 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "배경과 구도는 일치하나 캐릭터 조끼의 구조적 디테일이 심하게 누락되었습니다." } ] }, "combined": { "totals": { "A": 12, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S5sh1__bgfirst_bg.png", "bg_asset_id": "cbf7aa21-2e1d-4f36-91f6-f0bfc80c777c", "bg_record_key": "S5sh1::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S5sh4": { "input_fingerprint": "8857bb633e272085", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 타일 바닥에 맺힌 핏방울을 내려다보며 한 손으로 코를 꽉 틀어막고 있는 김지후(강수리영)의 놀란 표정.\n\nLOCATION (lock): The tiled floor area in the cramped service space behind the convenience-store refrigerator. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cramped refrigerator-back space, packed drink cans, white tile, and cool fluorescent light. Exclude the lunchbox emphasis and show only the worker and the fresh blood drop on the floor.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears the bone necklace, and the single dark-red drop of her nosebleed remains on the white tile floor because she has not wiped it away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 4 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "카메라를 실제 타일 바닥에 배치한 구도로, 핏방울을 내려다보며 코를 꽉 쥐고 있는 놀란 표정을 가장 사실적으로 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "코를 쥐고 있는 행동은 텍스트와 일치하나, 바닥이 아닌 가상의 타일 테이블을 만들어 공간 설정을 위반했습니다." }, { "label": "A", "score": 4, "verdict_ko": "코를 쥐지 않고 입을 가린 점이 아쉬우며, 공간에 존재하지 않는 타일 구조물을 임의로 생성해 배경이 왜곡되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S5sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "두 눈의 시선이 서로 다른 방향을 향하고 있으며 안구가 비정상적으로 돌출되어 있습니다.", "fix_en": "Realign both eyes to look consistently downward at the blood drop on the floor, and reduce the eye bulging to normal human proportions.", "unfixable": false }, { "issue_ko": "코를 막고 있는 손의 손가락들이 입술과 융합되어 기형적으로 묘사되었습니다.", "fix_en": "Redraw the hand on the face with correctly proportioned fingers pinching the nose, cleanly separated from the character's mouth.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Realign both eyes to look consistently downward at the blood drop on the floor, and reduce the eye bulging to normal human proportions.\n- Redraw the hand on the face with correctly proportioned fingers pinching the nose, cleanly separated from the character's mouth.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "우측 하단에 여분의 손이 나타나는 치명적인 신체 구조 오류(Hard Violation)가 발생했습니다." }, { "label": "B", "score": 3, "verdict_ko": "A와 시각적으로 동일한 이미지로, 마찬가지로 우측 하단에 여분의 손이 생성된 치명적인 오류가 있습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "우측 하단에 여분의 손이 나타나는 치명적인 신체 구조 오류(Hard Violation)가 발생했습니다." }, { "label": "B", "score": 3, "verdict_ko": "A와 시각적으로 동일한 이미지로, 마찬가지로 우측 하단에 여분의 손이 생성된 치명적인 오류가 있습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시문이 요구한 '코를 막고 바닥의 핏방울을 내려다보는 놀란 표정'과 복장, 목걸이 등 디테일을 참조 이미지에 맞게 매우 훌륭히 묘사했습니다." }, { "label": "B", "score": 8, "verdict_ko": "A와 시각적으로 동일하며, 지시문이 요구한 인물의 동작과 표정, 배경 및 연속성 요소를 모두 정확하게 반영했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시문이 요구한 '코를 막고 바닥의 핏방울을 내려다보는 놀란 표정'과 복장, 목걸이 등 디테일을 참조 이미지에 맞게 매우 훌륭히 묘사했습니다." }, { "label": "A", "score": 8, "verdict_ko": "A와 시각적으로 동일하며, 지시문이 요구한 인물의 동작과 표정, 배경 및 연속성 요소를 모두 정확하게 반영했습니다." } ] }, "combined": { "totals": { "A": 11, "B": 11 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S5sh1" } }, "S5sh8": { "input_fingerprint": "e44aa8f56938dc81", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 투명한 냉장고 유리문 너머에 서서 김지후(강수리영)를 향해 환하게 미소 짓고 있는 정인우의 얼굴.\n\nLOCATION (lock): Across the refrigerator’s glass door on the convenience-store sales floor, viewed from the rear service space. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same packed beverage refrigerator, narrow service space, and cool store lighting. Exclude the discarded food and blood-drop action, and place the smiling man only beyond the glass door.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears the bone necklace, and the unwiped blood drop remains on the white tile behind the refrigerator.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 5, "B": 3, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "핏방울과 인물의 표정을 잘 구현했으나, 공중에 뜬 문 아래로 다리가 없는 물리적 오류(Hard Violation)가 있습니다." }, { "label": "B", "score": 3, "verdict_ko": "명시적으로 금지된 인물 등장, 유령 반사상, 신체 절단 등 다수의 심각한 형태 및 지시 위반이 발생했습니다." }, { "label": "C", "score": 4, "verdict_ko": "인물의 표정은 적절하나 필수 요소인 핏방울이 누락되었고, 하반신이 생략된 물리적 오류가 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S5sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "유리문 프레임 아래로 이어지는 남자의 하반신과 다리가 없어 상체만 허공에 떠 있는 것처럼 보입니다.", "fix_en": "Render the man's lower torso and legs standing on the store floor in the empty space below the bottom edge of the glass door.", "unfixable": false }, { "issue_ko": "남자의 의상이 캐릭터 레퍼런스에 지정된 베이지색 코트와 회색 터틀넥이 아닙니다.", "fix_en": "Change the man's visible clothing to a grey turtleneck worn under a beige coat.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Render the man's lower torso and legs standing on the store floor in the empty space below the bottom edge of the glass door.\n- Change the man's visible clothing to a grey turtleneck worn under a beige coat.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "유리문 너머로 미소 짓는 인물의 얼굴과 바닥에 남은 핏방울을 지시대로 잘 구현했으나, 문 아래로 다리가 보이지 않는 렌더링 오류가 일부 존재합니다." }, { "label": "B", "score": 2, "verdict_ko": "문 아래에 기형적으로 축소된 다리가 생성되는 치명적인 신체/공간 구조 오류(Hard Violation)가 발생했으며, 유지되어야 할 바닥의 핏방울도 누락되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "유리문 너머로 미소 짓는 인물의 얼굴과 바닥에 남은 핏방울을 지시대로 잘 구현했으나, 문 아래로 다리가 보이지 않는 렌더링 오류가 일부 존재합니다." }, { "label": "B", "score": 2, "verdict_ko": "문 아래에 기형적으로 축소된 다리가 생성되는 치명적인 신체/공간 구조 오류(Hard Violation)가 발생했으며, 유지되어야 할 바닥의 핏방울도 누락되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "문 아래 공간에 하반신만 분리되어 나타나는 물리적으로 불가능한 신체 및 공간 구성 오류(Hard Violation)가 있으며 바닥의 핏방울도 누락되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "유리문 너머에 있는 인물의 얼굴을 사실적인 구도로 잘 담아내었으며, 타일 바닥에 남은 핏방울 등 요구된 디테일과 연속성을 충실히 반영했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 2, "verdict_ko": "문 아래 공간에 하반신만 분리되어 나타나는 물리적으로 불가능한 신체 및 공간 구성 오류(Hard Violation)가 있으며 바닥의 핏방울도 누락되었습니다." }, { "label": "A", "score": 7, "verdict_ko": "유리문 너머에 있는 인물의 얼굴을 사실적인 구도로 잘 담아내었으며, 타일 바닥에 남은 핏방울 등 요구된 디테일과 연속성을 충실히 반영했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 4 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S5sh4" } }, "S6sh1::bgfirst_bg": { "input_fingerprint": "e4451c997c1aee91", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 편의점 앞 벤치에 나란히 앉은 가운데 정인우가 김지후(강수리영)의 얼굴 앞까지 캔커피를 불쑥 내민 순간.\n\nLOCATION (lock): A sidewalk bench directly outside a convenience store at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From seated eye level directly before the bench, hold a medium two-person view as 정인우 thrusts the opened canned coffee close to 김지후(강수리영)'s face. The can sits near center without exceeding the figures in scale; 정인우 leans in from frame left while 김지후(강수리영) recoils slightly on frame right, keeping both seated bodies and the bench readable.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 정인우 in the middle-left of the frame, midground, reaches for 김지후 with canned coffee; 김지후(강수리영) in the middle-right of the frame, midground, looks toward offered canned coffee; canned coffee in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: bench (occupied by both characters) — Its front edge faces the camera with both characters seated side by side; used as Shared seating anchor that keeps both figures on a single confrontation line; canned coffee (opened and being offered) — Its cylindrical side angles toward the camera in 정인우's extended hand; used as Central handoff object positioned close to 김지후(강수리영)'s face; convenience store exterior (night exterior) — The store-facing side lies behind the seated pair; used as Establishes the exterior convenience-store location behind the bench.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 편의점 앞 벤치에 나란히 앉은 가운데 정인우가 김지후(강수리영)의 얼굴 앞까지 캔커피를 불쑥 내민 순간.\n\nLOCATION (lock): A sidewalk bench directly outside a convenience store at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From seated eye level directly before the bench, hold a medium two-person view as 정인우 thrusts the opened canned coffee close to 김지후(강수리영)'s face. The can sits near center without exceeding the figures in scale; 정인우 leans in from frame left while 김지후(강수리영) recoils slightly on frame right, keeping both seated bodies and the bench readable.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 정인우 in the middle-left of the frame, midground, reaches for 김지후 with canned coffee; 김지후(강수리영) in the middle-right of the frame, midground, looks toward offered canned coffee; canned coffee in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: bench (occupied by both characters) — Its front edge faces the camera with both characters seated side by side; used as Shared seating anchor that keeps both figures on a single confrontation line; canned coffee (opened and being offered) — Its cylindrical side angles toward the camera in 정인우's extended hand; used as Central handoff object positioned close to 김지후(강수리영)'s face; convenience store exterior (night exterior) — The store-facing side lies behind the seated pair; used as Establishes the exterior convenience-store location behind the bench.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S6sh1__bgfirst_bg.png", "asset_id": "bb521008-fd2f-4b9d-9758-014da40b8433", "input_asset_ids": [ "388c1280-6872-4715-90be-04e0e68404af", "ca44c205-ef26-47ee-9744-e98f1a1dfaaf" ] }, "S6sh1": { "input_fingerprint": "5acd750a45e5c129", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 편의점 앞 벤치에 나란히 앉은 가운데 정인우가 김지후(강수리영)의 얼굴 앞까지 캔커피를 불쑥 내민 순간.\n\nLOCATION (lock): A sidewalk bench directly outside a convenience store at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 편의점 앞 벤치에 나란히 앉은 가운데 정인우가 김지후(강수리영)의 얼굴 앞까지 캔커피를 불쑥 내민 순간.\n\nLOCATION (lock): A sidewalk bench directly outside a convenience store at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 편의점 앞 벤치에 나란히 앉은 가운데 정인우가 김지후(강수리영)의 얼굴 앞까지 캔커피를 불쑥 내민 순간.\n\nLOCATION (lock): A sidewalk bench directly outside a convenience store at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S6sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S6sh1.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L08B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "캔커피를 얼굴 바로 앞까지 불쑥 내민 행동과 놀란 표정을 정확하게 연출하여 제시된 상황을 가장 충실하게 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "캐릭터의 의상 일치도는 높으나, 캔커피를 건네는 거리가 멀어 '얼굴 앞까지 불쑥 내민' 핵심 행동 묘사를 살리지 못했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "캔커피를 얼굴 바로 앞까지 불쑥 내민 행동과 놀란 표정을 정확하게 연출하여 제시된 상황을 가장 충실하게 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "캐릭터의 의상 일치도는 높으나, 캔커피를 건네는 거리가 멀어 '얼굴 앞까지 불쑥 내민' 핵심 행동 묘사를 살리지 못했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지후의 얼굴 앞까지 캔커피를 불쑥 내미는 역동적인 동작과 동물 뼈 화살촉 목걸이의 디테일을 프롬프트에 맞게 훌륭하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "두 인물이 나란히 앉아 있지만, 캔커피를 지후의 얼굴 앞까지 내밀었다기보다는 둘 사이 허공에 든 것에 가깝고 목걸이의 형태도 다소 단순합니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지후의 얼굴 앞까지 캔커피를 불쑥 내미는 역동적인 동작과 동물 뼈 화살촉 목걸이의 디테일을 프롬프트에 맞게 훌륭하게 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "두 인물이 나란히 앉아 있지만, 캔커피를 지후의 얼굴 앞까지 내밀었다기보다는 둘 사이 허공에 든 것에 가깝고 목걸이의 형태도 다소 단순합니다." } ] }, "combined": { "totals": { "A": 12, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 12, "B": 16 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "캔커피를 얼굴 바로 앞까지 불쑥 내민 행동과 놀란 표정을 정확하게 연출하여 제시된 상황을 가장 충실하게 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "캐릭터의 의상 일치도는 높으나, 캔커피를 건네는 거리가 멀어 '얼굴 앞까지 불쑥 내민' 핵심 행동 묘사를 살리지 못했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L08B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "캔커피 표면에 텍스트와 로고가 선명하게 적혀 있음 (텍스트 금지 지침 위반).", "fix_en": "Remove all text, logos, and lettering from the blue coffee can, turning it into a completely plain, blank blue can.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text, logos, and lettering from the blue coffee can, turning it into a completely plain, blank blue can.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "두 인물의 배치, 캔커피를 건네는 동작, 지후의 뼈 목걸이 등 모든 샷 지침을 정확히 구현했으며 텍스트 금지 조건도 완벽히 준수했습니다." }, { "label": "A", "score": 6, "verdict_ko": "전반적인 연출과 소품 상태는 우수하나, 캔커피 표면에 상표 텍스트가 생성되어 '텍스트 없음' 지침을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "두 인물의 배치, 캔커피를 건네는 동작, 지후의 뼈 목걸이 등 모든 샷 지침을 정확히 구현했으며 텍스트 금지 조건도 완벽히 준수했습니다." }, { "label": "A", "score": 6, "verdict_ko": "전반적인 연출과 소품 상태는 우수하나, 캔커피 표면에 상표 텍스트가 생성되어 '텍스트 없음' 지침을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "텍스트 금지 조건을 준수하여 캔을 무지로 표현했으며, 벤치에서의 행동 묘사와 인물의 특징, 목걸이 착용 상태를 모두 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "인물, 배경, 연출 등은 A와 동일하게 훌륭하나, 캔 표면에 텍스트가 생성되어 '화면 내 텍스트 금지' 지시를 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "텍스트 금지 조건을 준수하여 캔을 무지로 표현했으며, 벤치에서의 행동 묘사와 인물의 특징, 목걸이 착용 상태를 모두 정확하게 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "인물, 배경, 연출 등은 A와 동일하게 훌륭하나, 캔 표면에 텍스트가 생성되어 '화면 내 텍스트 금지' 지시를 위반했습니다." } ] }, "combined": { "totals": { "A": 12, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S6sh1__bgfirst_bg.png", "bg_asset_id": "bb521008-fd2f-4b9d-9758-014da40b8433", "bg_record_key": "S6sh1::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S6sh5": { "input_fingerprint": "72a2055d8760e4e2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정인우의 걱정스러운 시선을 마주한 채 한쪽 입꼬리를 비스듬히 올리며 코웃음 치는 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): The nighttime seating area around the bench in front of the convenience store. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same convenience-store frontage, bench placement, pavement, and night lighting. Exclude the coffee-handoff moment and frame only the woman's mocking reaction.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 8, "B": 5, "C": 4 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "이전 샷의 벤치 착석 상태와 인물 간 좌우 위치를 정확히 유지하며 한쪽 입꼬리를 올린 표정을 훌륭히 담아냈습니다." }, { "label": "B", "score": 5, "verdict_ko": "배경에 빈 벤치가 나타나고 인물의 좌우 위치가 반전되어 이전 샷의 물리적 연속성 유지에 실패했습니다." }, { "label": "C", "score": 4, "verdict_ko": "공간 연속성 오류와 더불어 배경 간판에 텍스트가 생성되어 절대 금지된 텍스트 배제 지침을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S6sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S6sh1" } }, "S6sh7": { "input_fingerprint": "ec3dafe37c5fa388", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 바지 주머니에 손을 반쯤 넣은 채 무언가 떠오른 듯 눈을 크게 뜬 정인우의 놀란 얼굴.\n\nLOCATION (lock): Beside the convenience-store bench on the adjoining nighttime sidewalk. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same bench, storefront glow, pavement, and nighttime look. Exclude the woman's reaction and the extended coffee can, focusing on the man recalling something.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 8, "B": 4, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "정인우의 레퍼런스 외모와 의상(베이지색 코트)을 정확히 구현했으며, 지시대로 여성을 프레임에서 제외하여 가장 충실합니다." }, { "label": "B", "score": 4, "verdict_ko": "얼굴은 일치하나 이전 샷에 등장한 여성의 의상(카키색 재킷과 후드)을 남성에게 잘못 입혔습니다." }, { "label": "C", "score": 3, "verdict_ko": "명시적으로 제외하라고 지시한 여성을 프레임에 포함시켰으며, 남성의 의상 색상도 레퍼런스와 일치하지 않습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S6sh5_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S6sh5" } }, "S7sh2::bgfirst_bg": { "input_fingerprint": "96dcfa82be45e2e8", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 금방이라도 눈물을 터뜨릴 듯 절박한 눈빛으로 정인우의 눈을 똑바로 올려다보는 강민숙(홍은미)의 붉어진 얼굴.\n\nLOCATION (lock): An open parking area immediately outside a police station in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 강민숙(홍은미)'s eye level and slightly on 정인우's side of the eyeline, finish the dolly-in with a tight close view of her reddened face angled upward toward him. Her face occupies the center-left while 정인우's soft shoulder edge on the right preserves the height relationship and the direction of her plea.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 강민숙(홍은미) in the middle-left of the frame, foreground, looks toward 정인우 at frame right; 정인우 in the middle-right of the frame, foreground, looks toward 강민숙 pleading.\n- KEY BACKGROUND ELEMENTS: police station parking area (daytime); used as A restrained background cue establishing the remembered police-station encounter.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 금방이라도 눈물을 터뜨릴 듯 절박한 눈빛으로 정인우의 눈을 똑바로 올려다보는 강민숙(홍은미)의 붉어진 얼굴.\n\nLOCATION (lock): An open parking area immediately outside a police station in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 강민숙(홍은미)'s eye level and slightly on 정인우's side of the eyeline, finish the dolly-in with a tight close view of her reddened face angled upward toward him. Her face occupies the center-left while 정인우's soft shoulder edge on the right preserves the height relationship and the direction of her plea.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 강민숙(홍은미) in the middle-left of the frame, foreground, looks toward 정인우 at frame right; 정인우 in the middle-right of the frame, foreground, looks toward 강민숙 pleading.\n- KEY BACKGROUND ELEMENTS: police station parking area (daytime); used as A restrained background cue establishing the remembered police-station encounter.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh2__bgfirst_bg.png", "asset_id": "5e958c6c-f98a-4e4c-aa77-d8351c1567ce", "input_asset_ids": [ "b631f7c8-52c8-4af0-8215-fc7a0005f764", "323b4703-9101-4ed8-894f-2b85995acaa9" ] }, "S7sh2": { "input_fingerprint": "63f6e6b93b97f0d2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 금방이라도 눈물을 터뜨릴 듯 절박한 눈빛으로 정인우의 눈을 똑바로 올려다보는 강민숙(홍은미)의 붉어진 얼굴.\n\nLOCATION (lock): An open parking area immediately outside a police station in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 금방이라도 눈물을 터뜨릴 듯 절박한 눈빛으로 정인우의 눈을 똑바로 올려다보는 강민숙(홍은미)의 붉어진 얼굴.\n\nLOCATION (lock): An open parking area immediately outside a police station in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 금방이라도 눈물을 터뜨릴 듯 절박한 눈빛으로 정인우의 눈을 똑바로 올려다보는 강민숙(홍은미)의 붉어진 얼굴.\n\nLOCATION (lock): An open parking area immediately outside a police station in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S7sh2.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L09B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "상대방의 눈을 똑바로 올려다보는 시선 처리와 금방이라도 울 듯한 붉어진 얼굴의 묘사가 지시문과 정확히 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "시선이 카메라 쪽을 향하고 있어 상대방의 눈을 바라보지 않으며, 이미 눈물을 흘리고 있어 묘사된 순간과 다릅니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "상대방의 눈을 똑바로 올려다보는 시선 처리와 금방이라도 울 듯한 붉어진 얼굴의 묘사가 지시문과 정확히 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "시선이 카메라 쪽을 향하고 있어 상대방의 눈을 바라보지 않으며, 이미 눈물을 흘리고 있어 묘사된 순간과 다릅니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "'붉어진 얼굴'과 '절박한 눈빛'의 감정 표현이 텍스트에 부합하며, 인물을 똑바로 올려다보는 시선이 정확히 연출되었습니다." }, { "label": "B", "score": 6, "verdict_ko": "상대방을 올려다보는 구도는 맞으나, '붉어진 얼굴'의 묘사가 A에 비해 약하고 시선이 덜 직접적입니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "'붉어진 얼굴'과 '절박한 눈빛'의 감정 표현이 텍스트에 부합하며, 인물을 똑바로 올려다보는 시선이 정확히 연출되었습니다." }, { "label": "A", "score": 6, "verdict_ko": "상대방을 올려다보는 구도는 맞으나, '붉어진 얼굴'의 묘사가 A에 비해 약하고 시선이 덜 직접적입니다." } ] }, "combined": { "totals": { "A": 13, "B": 12 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 13, "B": 12 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "상대방의 눈을 똑바로 올려다보는 시선 처리와 금방이라도 울 듯한 붉어진 얼굴의 묘사가 지시문과 정확히 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "시선이 카메라 쪽을 향하고 있어 상대방의 눈을 바라보지 않으며, 이미 눈물을 흘리고 있어 묘사된 순간과 다릅니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L09B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh2__bgfirst_bg.png", "bg_asset_id": "5e958c6c-f98a-4e4c-aa77-d8351c1567ce", "bg_record_key": "S7sh2::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S7sh6": { "input_fingerprint": "744a7e559548c573", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 정인우의 손바닥 위로 두툼한 흰색 돈봉투를 강하게 밀어붙여 맞닿은 순간의 강민숙(홍은미) 손.\n\nLOCATION (lock): Between parked vehicles in the police-station parking lot. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The thick money envelope is transferred into Inwoo's hand and remains in his possession through the end of the flashback.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 5, "B": 3, "C": 8 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "지정된 손 클로즈업이 아닌 인물의 상반신을 포함하는 넓은 프레이밍으로 연출되어 샷 텍스트의 의도를 정확히 살리지 못했습니다." }, { "label": "B", "score": 3, "verdict_ko": "프레이밍이 넓을 뿐만 아니라, 봉투를 받는 남성의 손가락이 6개로 렌더링되는 치명적인 해부학적 오류가 있습니다." }, { "label": "C", "score": 8, "verdict_ko": "돈봉투를 밀어붙이는 손과 손바닥의 순간을 정확한 인서트 샷(클로즈업)으로 완벽하게 구현했으며, 주차된 차량 사이라는 배경 조건도 충실히 반영했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두툼한 흰색 돈봉투: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S7sh2" } }, "S7sh9::bgfirst_bg": { "input_fingerprint": "a1dfdb08377ea034", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 한 손에 돈봉투를 든 채 등을 돌려 저만치 멀어진 강민숙(홍은미)의 뒷모습을 우두커니 서서 멍하니 바라보는 정인우의 전신.\n\nLOCATION (lock): A standing area in the police-station lot, looking across the open pavement toward its exit.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level behind and slightly to one side of 정인우, complete the dolly-out into a wide composition that holds his full body in the left foreground and 강민숙(홍은미) receding in the right background. His lowered shoulders and money-envelope hand remain readable, while the distance between them becomes the dominant graphic tension of the parking area.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정인우 in the lower-left of the frame, foreground, looks toward departing 강민숙; 강민숙(홍은미) in the middle-right of the frame, background, moves toward far side of parking area.\n- KEY BACKGROUND ELEMENTS: police station parking area (daytime); used as The open layout makes the growing separation between the figures legible; thick white money envelope (held by 정인우) — Its broad white face angles partly toward the camera beside 정인우's leg; used as Held low in 정인우's hand as the lingering evidence of the exchange; police station exterior (daytime exterior) — The outward-facing side sits beyond 강민숙(홍은미) in the background; used as Provides the location boundary behind the departing figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is organized around the police station exterior parking area, which serves as the open camera and performance zone. A vehicle approach meets one outer edge of the parking area so the parking facility remains connected to the surrounding access network. On the building-facing edge, a station-side pedestrian entrance directly connects the parking area to the police station interior. The vehicle approach and station entrance therefore occupy different edges of the same parking area, preserving an open central span for movement and views across the site.\n- parking area: police station exterior parking area\n- approach: vehicle approach to the police station parking area\n- entrance: station-side pedestrian entrance adjoining the parking area\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 한 손에 돈봉투를 든 채 등을 돌려 저만치 멀어진 강민숙(홍은미)의 뒷모습을 우두커니 서서 멍하니 바라보는 정인우의 전신.\n\nLOCATION (lock): A standing area in the police-station lot, looking across the open pavement toward its exit.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level behind and slightly to one side of 정인우, complete the dolly-out into a wide composition that holds his full body in the left foreground and 강민숙(홍은미) receding in the right background. His lowered shoulders and money-envelope hand remain readable, while the distance between them becomes the dominant graphic tension of the parking area.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정인우 in the lower-left of the frame, foreground, looks toward departing 강민숙; 강민숙(홍은미) in the middle-right of the frame, background, moves toward far side of parking area.\n- KEY BACKGROUND ELEMENTS: police station parking area (daytime); used as The open layout makes the growing separation between the figures legible; thick white money envelope (held by 정인우) — Its broad white face angles partly toward the camera beside 정인우's leg; used as Held low in 정인우's hand as the lingering evidence of the exchange; police station exterior (daytime exterior) — The outward-facing side sits beyond 강민숙(홍은미) in the background; used as Provides the location boundary behind the departing figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is organized around the police station exterior parking area, which serves as the open camera and performance zone. A vehicle approach meets one outer edge of the parking area so the parking facility remains connected to the surrounding access network. On the building-facing edge, a station-side pedestrian entrance directly connects the parking area to the police station interior. The vehicle approach and station entrance therefore occupy different edges of the same parking area, preserving an open central span for movement and views across the site.\n- parking area: police station exterior parking area\n- approach: vehicle approach to the police station parking area\n- entrance: station-side pedestrian entrance adjoining the parking area\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh9__bgfirst_bg.png", "asset_id": "595eaa1d-0f94-407c-9a4d-0ab5a850a7ed", "input_asset_ids": [ "e33916c3-32d9-4654-9970-08f9e12a2edc" ] }, "S7sh9": { "input_fingerprint": "ac40115383ec888d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한 손에 돈봉투를 든 채 등을 돌려 저만치 멀어진 강민숙(홍은미)의 뒷모습을 우두커니 서서 멍하니 바라보는 정인우의 전신.\n\nLOCATION (lock): A standing area in the police-station lot, looking across the open pavement toward its exit. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo remains holding the thick money envelope as he watches Minsook walk away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한 손에 돈봉투를 든 채 등을 돌려 저만치 멀어진 강민숙(홍은미)의 뒷모습을 우두커니 서서 멍하니 바라보는 정인우의 전신.\n\nLOCATION (lock): A standing area in the police-station lot, looking across the open pavement toward its exit. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo remains holding the thick money envelope as he watches Minsook walk away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한 손에 돈봉투를 든 채 등을 돌려 저만치 멀어진 강민숙(홍은미)의 뒷모습을 우두커니 서서 멍하니 바라보는 정인우의 전신.\n\nLOCATION (lock): A standing area in the police-station lot, looking across the open pavement toward its exit. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo remains holding the thick money envelope as he watches Minsook walk away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한 손에 돈봉투를 든 채 등을 돌려 저만치 멀어진 강민숙(홍은미)의 뒷모습을 우두커니 서서 멍하니 바라보는 정인우의 전신.\n\nLOCATION (lock): A standing area in the police-station lot, looking across the open pavement toward its exit. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo remains holding the thick money envelope as he watches Minsook walk away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh9__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두툼한 흰색 돈봉투: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh9__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두툼한 흰색 돈봉투: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh9__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두툼한 흰색 돈봉투: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "totals": { "A": 8, "C": 7, "B": 5 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "프롬프트에 명시된 '단정한 짙은색 짧은 머리'를 정확히 반영했으며, 전신 구도와 인물 간의 거리, 봉투를 쥔 손의 형태가 가장 자연스럽습니다." }, { "label": "C", "score": 7, "verdict_ko": "지문 내용을 전반적으로 잘 따랐으나, 봉투를 쥔 오른손의 손가락 구조가 다소 어색하게 왜곡되어 A보다 완성도가 낮습니다." }, { "label": "B", "score": 5, "verdict_ko": "지문에 '단정한 짙은색 짧은 머리'로 묘사되었음에도 레퍼런스의 비니를 그대로 착용한 채 생성되어 캐릭터 외형 지시를 위반했습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh9__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두툼한 흰색 돈봉투: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S7sh9__bgfirst_bg.png", "bg_asset_id": "595eaa1d-0f94-407c-9a4d-0ab5a850a7ed", "bg_record_key": "S7sh9::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S7sh6" } }, "S8sh2": { "input_fingerprint": "1ea7eb83aa9bbce3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 빼앗은 돈봉투를 쥔 손을 허공에 치켜든 채 잔뜩 화가 난 눈빛으로 정인우를 노려보는 김지후(강수리영)의 상체.\n\nLOCATION (lock): The sidewalk beside the bench outside the convenience store at night. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same convenience-store frontage, bench, pavement, and night illumination. Exclude the earlier coffee handoff and show the woman holding the returned envelope while confronting the man.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her bone necklace and now possesses the same thick money envelope she has just snatched from Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 8, "B": 6, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "지시된 상체 샷과 감정 연기를 잘 구현했으며, 레퍼런스의 독특한 돈봉투 형태(덮개 홈)를 완벽하게 재현함." }, { "label": "B", "score": 6, "verdict_ko": "프레이밍과 배경은 적절하나 돈봉투의 형태가 레퍼런스와 다르고, 봉투를 쥔 손가락의 묘사가 다소 어색함." }, { "label": "A", "score": 3, "verdict_ko": "허용된 인물(김지후) 외에 타인의 뒷모습을 화면에 포함하여 엄격한 인물 제한 규정을 위반함." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S6sh7_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두툼한 흰색 돈봉투: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S6sh7" } }, "S8sh8": { "input_fingerprint": "a950e9a134a5d572", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 무언가 불길한 사실을 깨달은 듯 입술을 굳게 다문 채 딱딱하게 굳어버린 김지후(강수리영)의 창백한 표정.\n\nLOCATION (lock): The nighttime bench area directly in front of the convenience store. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same storefront glow, bench position, and dark pavement. Exclude the coffee-handoff action and frame only the woman's suddenly rigid realization.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to hold or retain the thick money envelope and still wears her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "C": 6, "A": 4 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "얼굴에 집중한 타이트한 프레이밍으로 이전 샷의 동작을 배제하라는 지시를 잘 따랐으며, 창백하게 굳은 표정을 가장 정확하게 포착했습니다." }, { "label": "C", "score": 6, "verdict_ko": "표정 묘사와 배경은 준수하나, 위로 들고 있어야 할 팔을 아래로 내린 모습이 프레임에 포함되어 자세 연속성 유지 지시를 위반했습니다." }, { "label": "A", "score": 4, "verdict_ko": "팔의 자세가 임의로 변경되어 프레임에 나타났으며, 배경의 벤치 위치가 반대로 뒤집혀 장소의 물리적 연속성 오류가 발생했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S8sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S8sh2" } }, "S9sh2::bgfirst_bg": { "input_fingerprint": "335d3f55ba265f20", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 소파 쪽을 바라보며 두 눈이 커진 채 얼어붙은 강민숙(홍은미)의 하얗게 질린 얼굴.\n\nLOCATION (lock): The small living room of a modest two-room multi-family home, viewed from the entryway under indoor nighttime light.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the sofa side at eye level, the inward move ends with 강민숙(홍은미)'s face occupying most of the frame while her body remains angled toward the unseen seated woman. Her enlarged eyes are held above center and the edge of one raised shoulder compresses the lower frame, preserving the terrified eyeline toward the sofa without revealing its occupant.\n- FRAMING SCALE: close-up\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 소파 쪽을 바라보며 두 눈이 커진 채 얼어붙은 강민숙(홍은미)의 하얗게 질린 얼굴.\n\nLOCATION (lock): The small living room of a modest two-room multi-family home, viewed from the entryway under indoor nighttime light.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the sofa side at eye level, the inward move ends with 강민숙(홍은미)'s face occupying most of the frame while her body remains angled toward the unseen seated woman. Her enlarged eyes are held above center and the edge of one raised shoulder compresses the lower frame, preserving the terrified eyeline toward the sofa without revealing its occupant.\n- FRAMING SCALE: close-up\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S9sh2__bgfirst_bg.png", "asset_id": "9602c378-d1e8-4ee5-bb2a-d0632e341c2b", "input_asset_ids": [ "8540f759-f4d9-4bfa-a4a9-c4140ecbb688", "ab516e62-d46b-4428-a8a6-07f81dd47697" ] }, "S9sh2": { "input_fingerprint": "3a003b79f09eb7a8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 소파 쪽을 바라보며 두 눈이 커진 채 얼어붙은 강민숙(홍은미)의 하얗게 질린 얼굴.\n\nLOCATION (lock): The small living room of a modest two-room multi-family home, viewed from the entryway under indoor nighttime light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 소파 쪽을 바라보며 두 눈이 커진 채 얼어붙은 강민숙(홍은미)의 하얗게 질린 얼굴.\n\nLOCATION (lock): The small living room of a modest two-room multi-family home, viewed from the entryway under indoor nighttime light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 소파 쪽을 바라보며 두 눈이 커진 채 얼어붙은 강민숙(홍은미)의 하얗게 질린 얼굴.\n\nLOCATION (lock): The small living room of a modest two-room multi-family home, viewed from the entryway under indoor nighttime light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S9sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S9sh2.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L10B02.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'얼굴'에 집중한 프레이밍과 소파를 향한 시선, 크게 뜬 두 눈의 묘사가 지문의 의도를 훌륭하게 반영합니다." }, { "label": "B", "score": 4, "verdict_ko": "'얼굴'을 강조한 지문과 달리 프레임이 너무 넓으며, 소파를 바라보는 시선 처리도 명확하지 않습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'얼굴'에 집중한 프레이밍과 소파를 향한 시선, 크게 뜬 두 눈의 묘사가 지문의 의도를 훌륭하게 반영합니다." }, { "label": "B", "score": 4, "verdict_ko": "'얼굴'을 강조한 지문과 달리 프레임이 너무 넓으며, 소파를 바라보는 시선 처리도 명확하지 않습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "'하얗게 질린 얼굴'과 '두 눈이 커진 채'라는 핵심 지문을 요구된 프레이밍(얼굴 중심의 샷)으로 정확하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "인물의 얼굴 표정에 집중해야 하는 지문임에도 풀샷으로 촬영하여 샷의 프레이밍과 핵심 연출 의도를 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'하얗게 질린 얼굴'과 '두 눈이 커진 채'라는 핵심 지문을 요구된 프레이밍(얼굴 중심의 샷)으로 정확하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "인물의 얼굴 표정에 집중해야 하는 지문임에도 풀샷으로 촬영하여 샷의 프레이밍과 핵심 연출 의도를 위반했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 7 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 7 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'얼굴'에 집중한 프레이밍과 소파를 향한 시선, 크게 뜬 두 눈의 묘사가 지문의 의도를 훌륭하게 반영합니다." }, { "label": "B", "score": 4, "verdict_ko": "'얼굴'을 강조한 지문과 달리 프레임이 너무 넓으며, 소파를 바라보는 시선 처리도 명확하지 않습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L10B02.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S9sh2__bgfirst_bg.png", "bg_asset_id": "9602c378-d1e8-4ee5-bb2a-d0632e341c2b", "bg_record_key": "S9sh2::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S9sh6": { "input_fingerprint": "226552db74f73d39", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 현관문 앞을 떡 하니 막아선 엄청난 덩치의 황봉업(황 씨·민숙 집의 거구 남자)의 산 같은 뒷모습.\n\nLOCATION (lock): The interior entryway of the multi-family home, with the front door blocked by a large figure. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 4, "B": 7, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "캐릭터의 의상(반사띠가 없는 단순한 재킷)이 레퍼런스와 일치하지 않으며, 현관문이 아닌 거실 쪽을 향해 서 있어 연출이 아쉽습니다." }, { "label": "B", "score": 7, "verdict_ko": "캐릭터의 특징적인 점퍼 의상이 정확히 반영되었으며, 어두운 현관문을 막아선 거대한 뒷모습과 이전 샷의 거실 배경이 훌륭하게 구현되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "캐릭터의 의상이 레퍼런스와 다르고, 지정된 거실 배경의 디테일이 화면에 전혀 나타나지 않습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S9sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S9sh2" } }, "S10sh2::bgfirst_bg": { "input_fingerprint": "1fabdcb8747728fc", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 두툼한 흰색 돈봉투를 손가락으로 꽉 쥐고 있는 김지후(강수리영)의 양손 클로즈업.\n\nLOCATION (lock): Behind the checkout counter on the convenience-store sales floor at night, under the store’s active lighting.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Immediately across the counter at hand height, the static insert isolates 김지후(강수리영)'s two hands crushing the thick white money envelope, while a narrow strip of counter and her forearms retain scale and location. The envelope remains below forty percent of the image, angled across the lower center as whitening knuckles and compressed fingers carry the tension.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: thick white money envelope (tightly gripped) — Its broad face is angled toward the camera while the compressed edge shows its thickness; used as Primary handled prop held between both hands in the lower-center of frame; convenience-store counter (present); used as Provides a narrow spatial reference beneath the hands.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 두툼한 흰색 돈봉투를 손가락으로 꽉 쥐고 있는 김지후(강수리영)의 양손 클로즈업.\n\nLOCATION (lock): Behind the checkout counter on the convenience-store sales floor at night, under the store’s active lighting.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Immediately across the counter at hand height, the static insert isolates 김지후(강수리영)'s two hands crushing the thick white money envelope, while a narrow strip of counter and her forearms retain scale and location. The envelope remains below forty percent of the image, angled across the lower center as whitening knuckles and compressed fingers carry the tension.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: thick white money envelope (tightly gripped) — Its broad face is angled toward the camera while the compressed edge shows its thickness; used as Primary handled prop held between both hands in the lower-center of frame; convenience-store counter (present); used as Provides a narrow spatial reference beneath the hands.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S10sh2__bgfirst_bg.png", "asset_id": "540afd0b-5c6e-47cb-9297-b162fe00b7b3", "input_asset_ids": [ "af6b6d5c-a1fe-4cc2-a494-933f4fa5dffa", "c36b40bf-6408-4a31-b800-fc6912c8be0d" ] }, "S10sh2": { "input_fingerprint": "d357fcbd2f38dc7f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 두툼한 흰색 돈봉투를 손가락으로 꽉 쥐고 있는 김지후(강수리영)의 양손 클로즈업.\n\nLOCATION (lock): Behind the checkout counter on the convenience-store sales floor at night, under the store’s active lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her bone necklace and continues handling the same thick money envelope taken from Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 두툼한 흰색 돈봉투를 손가락으로 꽉 쥐고 있는 김지후(강수리영)의 양손 클로즈업.\n\nLOCATION (lock): Behind the checkout counter on the convenience-store sales floor at night, under the store’s active lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her bone necklace and continues handling the same thick money envelope taken from Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 두툼한 흰색 돈봉투를 손가락으로 꽉 쥐고 있는 김지후(강수리영)의 양손 클로즈업.\n\nLOCATION (lock): Behind the checkout counter on the convenience-store sales floor at night, under the store’s active lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her bone necklace and continues handling the same thick money envelope taken from Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S10sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S10sh2.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두툼한 흰색 돈봉투: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L07B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두툼한 흰색 돈봉투: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "양손 클로즈업 프레이밍을 유지하면서 돈봉투를 꽉 쥐는 행동과 뼈 목걸이의 올바른 착용 위치를 자연스럽게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 뼈 목걸이를 손목에 차는 오류를 범했으며, 봉투를 꽉 쥐는 묘사가 약하고 맨팔로 나타나 의상 설정과 어긋납니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "양손 클로즈업 프레이밍을 유지하면서 돈봉투를 꽉 쥐는 행동과 뼈 목걸이의 올바른 착용 위치를 자연스럽게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 뼈 목걸이를 손목에 차는 오류를 범했으며, 봉투를 꽉 쥐는 묘사가 약하고 맨팔로 나타나 의상 설정과 어긋납니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "양손 클로즈업, 돈봉투 쥐기, 뼈 목걸이 착용 위치 등 지시사항을 충실히 따랐으며 장소를 입체적으로 잘 활용했습니다." }, { "label": "B", "score": 3, "verdict_ko": "장소 레퍼런스의 카메라 구도를 그대로 복사하지 말라는 지시를 위반했으며, 목걸이를 손목에 착용하는 오류가 발생했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "양손 클로즈업, 돈봉투 쥐기, 뼈 목걸이 착용 위치 등 지시사항을 충실히 따랐으며 장소를 입체적으로 잘 활용했습니다." }, { "label": "A", "score": 3, "verdict_ko": "장소 레퍼런스의 카메라 구도를 그대로 복사하지 말라는 지시를 위반했으며, 목걸이를 손목에 착용하는 오류가 발생했습니다." } ] }, "combined": { "totals": { "A": 7, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 7, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "양손 클로즈업 프레이밍을 유지하면서 돈봉투를 꽉 쥐는 행동과 뼈 목걸이의 올바른 착용 위치를 자연스럽게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 뼈 목걸이를 손목에 차는 오류를 범했으며, 봉투를 꽉 쥐는 묘사가 약하고 맨팔로 나타나 의상 설정과 어긋납니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L07B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두툼한 흰색 돈봉투: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S10sh2__bgfirst_bg.png", "bg_asset_id": "540afd0b-5c6e-47cb-9297-b162fe00b7b3", "bg_record_key": "S10sh2::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S10sh5": { "input_fingerprint": "d2ee975f1f339cda", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 핸드폰 화면에 떠오른 '도망쳐'라는 메시지를 보며 어깨를 한껏 움츠린 채 두 눈을 크게 뜬 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): The convenience-store checkout counter, illuminated by the phone screen and the store’s nighttime lights. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The money envelope remains with Jihu at the counter, and she still wears her bone necklace while reading the message.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 3, "B": 2 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "화면과 텍스트가 캐릭터의 시점을 향해 소품 방향 규칙을 완벽히 준수했으며, 움츠린 어깨와 크게 뜬 눈의 연기도 매우 정확합니다." }, { "label": "A", "score": 3, "verdict_ko": "캐릭터가 화면을 읽고 있음에도 핸드폰 정면이 카메라를 향하고 있어 소품 방향 규칙을 위반한 심각한 오류가 있습니다." }, { "label": "B", "score": 2, "verdict_ko": "핸드폰 화면이 카메라를 향해 소품 규칙을 위반했으며, 어깨를 한껏 움츠린 자세 지시도 제대로 반영되지 않았습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S10sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S10sh2" } }, "S11sh3::bgfirst_bg": { "input_fingerprint": "124c4ec3a9e26298", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 활짝 열린 경찰차 운전석 문밖으로 다급한 표정을 한 채 몸을 한껏 내밀고 있는 정인우의 상체.\n\nLOCATION (lock): A narrow residential alley at the curb beside a stopped police car at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Just outside the fully open driver's door at upper-body height, the dolly-in settles on 정인우 leaning urgently through the doorway, his torso isolated beyond the angled door edge. He occupies the center-right of the frame in three-quarter view, while enough of the car body and roofline remains visible to establish where he has emerged from.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: driver's door (fully open) — The door is fully open, with its inner side angled toward the camera; used as Creates a hard foreground edge around Inwoo's emerging upper body; police car (stopped) — The driver's side and roofline face the camera at an oblique angle; used as Provides vehicle context behind and below the open door; red warning beacon (mounted; operating state unspecified) — Mounted on the police-car roof with its side visible to camera; used as A restrained red visual marker along the visible roofline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 활짝 열린 경찰차 운전석 문밖으로 다급한 표정을 한 채 몸을 한껏 내밀고 있는 정인우의 상체.\n\nLOCATION (lock): A narrow residential alley at the curb beside a stopped police car at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Just outside the fully open driver's door at upper-body height, the dolly-in settles on 정인우 leaning urgently through the doorway, his torso isolated beyond the angled door edge. He occupies the center-right of the frame in three-quarter view, while enough of the car body and roofline remains visible to establish where he has emerged from.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: driver's door (fully open) — The door is fully open, with its inner side angled toward the camera; used as Creates a hard foreground edge around Inwoo's emerging upper body; police car (stopped) — The driver's side and roofline face the camera at an oblique angle; used as Provides vehicle context behind and below the open door; red warning beacon (mounted; operating state unspecified) — Mounted on the police-car roof with its side visible to camera; used as A restrained red visual marker along the visible roofline.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S11sh3__bgfirst_bg.png", "asset_id": "b1043950-0c2b-49f7-a19b-f6a8c62bbb6d", "input_asset_ids": [ "5403ac85-c5cf-4b23-a167-d91df3619c15", "47021160-eed2-4642-9cc1-2964d4c92934" ] }, "S11sh3": { "input_fingerprint": "90c1dac6079ccf59", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 활짝 열린 경찰차 운전석 문밖으로 다급한 표정을 한 채 몸을 한껏 내밀고 있는 정인우의 상체.\n\nLOCATION (lock): A narrow residential alley at the curb beside a stopped police car at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 활짝 열린 경찰차 운전석 문밖으로 다급한 표정을 한 채 몸을 한껏 내밀고 있는 정인우의 상체.\n\nLOCATION (lock): A narrow residential alley at the curb beside a stopped police car at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 활짝 열린 경찰차 운전석 문밖으로 다급한 표정을 한 채 몸을 한껏 내밀고 있는 정인우의 상체.\n\nLOCATION (lock): A narrow residential alley at the curb beside a stopped police car at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S11sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S11sh3.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L11B02.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "차량 창문에 프롬프트에 없는 인물의 얼굴이 반사되어 나타나며, 신체의 팔 부위가 생략된 치명적인 해부학적 오류(Hard Violation)가 있습니다." }, { "label": "B", "score": 6, "verdict_ko": "운전석 안에서 밖으로 몸을 내미는 연출 대신 이미 차 밖에 서 있는 모습으로 그려졌고 레퍼런스의 의상이 경찰복으로 변경되었으나, 배경과 인물 외형은 잘 반영되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "차량 창문에 프롬프트에 없는 인물의 얼굴이 반사되어 나타나며, 신체의 팔 부위가 생략된 치명적인 해부학적 오류(Hard Violation)가 있습니다." }, { "label": "B", "score": 6, "verdict_ko": "운전석 안에서 밖으로 몸을 내미는 연출 대신 이미 차 밖에 서 있는 모습으로 그려졌고 레퍼런스의 의상이 경찰복으로 변경되었으나, 배경과 인물 외형은 잘 반영되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "운전석 안에서 문밖으로 상체를 다급하게 내미는 핵심 행동과 프레이밍을 지시문에 맞게 매우 정확히 연출했습니다." }, { "label": "A", "score": 4, "verdict_ko": "문밖으로 몸을 내미는 대신 이미 차 밖에 서 있으며, 차량 옆면에 불필요한 텍스트가 생성되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "운전석 안에서 문밖으로 상체를 다급하게 내미는 핵심 행동과 프레이밍을 지시문에 맞게 매우 정확히 연출했습니다." }, { "label": "B", "score": 4, "verdict_ko": "문밖으로 몸을 내미는 대신 이미 차 밖에 서 있으며, 차량 옆면에 불필요한 텍스트가 생성되었습니다." } ] }, "combined": { "totals": { "A": 10, "B": 10 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 10, "B": 10 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "차량 창문에 프롬프트에 없는 인물의 얼굴이 반사되어 나타나며, 신체의 팔 부위가 생략된 치명적인 해부학적 오류(Hard Violation)가 있습니다." }, { "label": "B", "score": 6, "verdict_ko": "운전석 안에서 밖으로 몸을 내미는 연출 대신 이미 차 밖에 서 있는 모습으로 그려졌고 레퍼런스의 의상이 경찰복으로 변경되었으나, 배경과 인물 외형은 잘 반영되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L11B02.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S11sh3__bgfirst_bg.png", "bg_asset_id": "b1043950-0c2b-49f7-a19b-f6a8c62bbb6d", "bg_record_key": "S11sh3::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S11sh4": { "input_fingerprint": "a1c3b39f6c189f7a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 다세대 주택의 낡은 외부 계단에서 한 발은 계단을 딛고 다른 발은 허공에 뜬 mid-action 자세로 뛰어 올라가는 정인우와 김지후(강수리영)의 뒷모습.\n\nLOCATION (lock): The exposed exterior stairway climbing the side of a multi-family residential building. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to carry her phone in hand and still wears the arrowhead-shaped bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "한 발이 허공에 뜬 달리는 뒷모습을 역동적으로 완벽히 구현했으며, 이전 샷의 복장(코트 없음) 및 장소 설정도 정확히 유지했습니다." }, { "label": "A", "score": 5, "verdict_ko": "배경과 인물 외형은 무난하나, '한 발이 허공에 뜬 달리는 자세'가 아닌 양발이 계단에 닿아 있는 걷는 모습으로 연출되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "남성의 왼쪽 발굽 아래로 기둥이 생겨 계단과 융합되는 치명적인 신체 구조 오류(Hard Violation)가 발생했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S11sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S11sh3" }, "lane_policy": "ab_select_bypass:prev" }, "S12sh3::bgfirst_bg": { "input_fingerprint": "882242f307a42d6d", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 피 묻은 난장판 한가운데 서서 부서진 사진 액자를 손에 든 채 서 있는 날카로운 눈매의 고산들의 전신.\n\nLOCATION (lock): The wrecked central living area of the multi-family home, among broken furnishings and visible bloodstains.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Just inside the entrance at eye level, a static wide composition places 고산들 full-length in the central depth of the room, isolated by the surrounding destruction. His body faces the entrance while his head and broken photograph frame turn slightly off-axis, allowing the damaged frame and his cold stillness to register without flattening him into a frontal portrait.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: broken photograph frame (broken) — The photograph-bearing face is visible obliquely to camera, but its image remains indistinct; the damaged frame edge is evident; used as Held at Sandeul's torso as the key narrative prop within the wider reveal; broken furniture (broken and disordered) — Broken forms point at varied angles around Sandeul, leaving him readable in the central depth; used as Surrounds the centered figure and establishes the violent disorder; broken lamp (broken) — Its damaged body is visible among the displaced room contents; used as Adds evidence of destruction within the room around the figure; broken flowerpot (broken) — Its broken form lies among the surrounding disorder; used as Provides scattered evidence of the ransacked interior; blood marks (clearly visible in multiple areas); used as Creates visible evidence of violence around the central figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 피 묻은 난장판 한가운데 서서 부서진 사진 액자를 손에 든 채 서 있는 날카로운 눈매의 고산들의 전신.\n\nLOCATION (lock): The wrecked central living area of the multi-family home, among broken furnishings and visible bloodstains.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Just inside the entrance at eye level, a static wide composition places 고산들 full-length in the central depth of the room, isolated by the surrounding destruction. His body faces the entrance while his head and broken photograph frame turn slightly off-axis, allowing the damaged frame and his cold stillness to register without flattening him into a frontal portrait.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: broken photograph frame (broken) — The photograph-bearing face is visible obliquely to camera, but its image remains indistinct; the damaged frame edge is evident; used as Held at Sandeul's torso as the key narrative prop within the wider reveal; broken furniture (broken and disordered) — Broken forms point at varied angles around Sandeul, leaving him readable in the central depth; used as Surrounds the centered figure and establishes the violent disorder; broken lamp (broken) — Its damaged body is visible among the displaced room contents; used as Adds evidence of destruction within the room around the figure; broken flowerpot (broken) — Its broken form lies among the surrounding disorder; used as Provides scattered evidence of the ransacked interior; blood marks (clearly visible in multiple areas); used as Creates visible evidence of violence around the central figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S12sh3__bgfirst_bg.png", "asset_id": "0e3379b1-847c-4d84-ac74-5dcea8b5acf5", "input_asset_ids": [ "b2a14138-70e4-489f-9899-ae8d4fca9bbb", "84d273d8-a577-4413-998c-1d0cc640befb" ] }, "S12sh3": { "input_fingerprint": "02be563469cd4b8d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 피 묻은 난장판 한가운데 서서 부서진 사진 액자를 손에 든 채 서 있는 날카로운 눈매의 고산들의 전신.\n\nLOCATION (lock): The wrecked central living area of the multi-family home, among broken furnishings and visible bloodstains. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul is holding the photograph frame taken from Minsook's wrecked home; the broken furniture, shattered objects, and bloodstains remain throughout the room.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 피 묻은 난장판 한가운데 서서 부서진 사진 액자를 손에 든 채 서 있는 날카로운 눈매의 고산들의 전신.\n\nLOCATION (lock): The wrecked central living area of the multi-family home, among broken furnishings and visible bloodstains. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul is holding the photograph frame taken from Minsook's wrecked home; the broken furniture, shattered objects, and bloodstains remain throughout the room.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 피 묻은 난장판 한가운데 서서 부서진 사진 액자를 손에 든 채 서 있는 날카로운 눈매의 고산들의 전신.\n\nLOCATION (lock): The wrecked central living area of the multi-family home, among broken furnishings and visible bloodstains. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul is holding the photograph frame taken from Minsook's wrecked home; the broken furniture, shattered objects, and bloodstains remain throughout the room.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S12sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S12sh3.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 낡은 사진 액자: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L10B01.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 낡은 사진 액자: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '전신' 프레이밍을 정확히 구현했으며, 규칙에 맞게 액자의 앞면을 캐릭터의 시선 쪽으로 향하게 연출해 프롬프트 충실도가 높습니다." }, { "label": "B", "score": 4, "verdict_ko": "'전신' 샷 지시를 위반하여 하반신이 프레임에서 잘렸으며, 명시적인 지시 없이 액자 앞면을 카메라 쪽으로 향하게 한 소품 규칙 위반이 있습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '전신' 프레이밍을 정확히 구현했으며, 규칙에 맞게 액자의 앞면을 캐릭터의 시선 쪽으로 향하게 연출해 프롬프트 충실도가 높습니다." }, { "label": "B", "score": 4, "verdict_ko": "'전신' 샷 지시를 위반하여 하반신이 프레임에서 잘렸으며, 명시적인 지시 없이 액자 앞면을 카메라 쪽으로 향하게 한 소품 규칙 위반이 있습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "전신(풀샷) 앵글 요구사항을 어기고 피사체를 가깝게 잡았으며, 지정된 배경 레퍼런스의 공간 구조를 임의로 변경하여 위치 조건을 위반했습니다." }, { "label": "B", "score": 8, "verdict_ko": "지정된 텍스트대로 전신 샷의 구도를 정확히 구현했으며, 고정된 배경 레퍼런스의 시점과 구조를 훌륭하게 유지했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 3, "verdict_ko": "전신(풀샷) 앵글 요구사항을 어기고 피사체를 가깝게 잡았으며, 지정된 배경 레퍼런스의 공간 구조를 임의로 변경하여 위치 조건을 위반했습니다." }, { "label": "A", "score": 8, "verdict_ko": "지정된 텍스트대로 전신 샷의 구도를 정확히 구현했으며, 고정된 배경 레퍼런스의 시점과 구조를 훌륭하게 유지했습니다." } ] }, "combined": { "totals": { "A": 15, "B": 7 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 15, "B": 7 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '전신' 프레이밍을 정확히 구현했으며, 규칙에 맞게 액자의 앞면을 캐릭터의 시선 쪽으로 향하게 연출해 프롬프트 충실도가 높습니다." }, { "label": "B", "score": 4, "verdict_ko": "'전신' 샷 지시를 위반하여 하반신이 프레임에서 잘렸으며, 명시적인 지시 없이 액자 앞면을 카메라 쪽으로 향하게 한 소품 규칙 위반이 있습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L10B01.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 낡은 사진 액자: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S12sh3__bgfirst_bg.png", "bg_asset_id": "0e3379b1-847c-4d84-ac74-5dcea8b5acf5", "bg_record_key": "S12sh3::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S12sh9": { "input_fingerprint": "8bc88f231b401e69", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 고산들의 가슴팍에 손을 맞대고 강하게 밀어내는 도중, 힘이 팽팽하게 실린 정인우의 어깨와 뒤로 젖혀지려는 고산들의 상체.\n\nLOCATION (lock): The damaged living-room floor between the intruder and the home’s entrance. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same wrecked living room, broken furniture, blood traces, and harsh interior lighting. Exclude the broken picture-frame pose and preserve only the two men struggling in the room.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wrecked furniture, broken fixtures, and visible bloodstains remain fixed throughout Minsook's home, and Jihu still wears her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 4, "A": 2 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "정인우가 고산들을 밀쳐내는 역동적인 자세와 이전 샷의 조명, 고산들의 복장을 잘 유지했으나, 정인우의 코트 색상이 기준과 달라 감점되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "두 인물의 물리적 충돌은 묘사되었으나, 이전 샷의 실내 조명 톤을 무시하고 푸른빛으로 변경되었으며 고산들의 셔츠 색상도 일치하지 않습니다." }, { "label": "A", "score": 2, "verdict_ko": "두 인물의 역할(밀어내는 자와 밀리는 자)이 뒤바뀌었고, 프롬프트에 없는 부적절한 민소매 복장이 등장하여 심각한 오류가 발생했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S12sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S12sh3" } }, "S12sh16": { "input_fingerprint": "b195c8ae39e2cbb2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 바닥에 엎드린 채 사색이 되어 숨을 죽이고 위를 올려다보는 김지후(강수리영)의 공포에 질린 표정.\n\nLOCATION (lock): On the floor of the ransacked living room, amid broken household objects and blood marks. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same destroyed living room, blood stains, debris, and damaged walls. Exclude the standing man with the frame and show the terrified woman prone on the floor.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains on the floor where Sandeul threw her, still wearing her bone necklace; Inwoo remains down by the wall after being hurled into it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 5 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "바닥에 엎드려 위를 보는 공포스러운 표정과 지정된 복장(재킷, 목걸이)을 잘 구현했으며, 불필요한 인물을 정확히 배제함." }, { "label": "B", "score": 3, "verdict_ko": "명시적으로 제외하라고 지시한 서 있는 남성을 배경에 포함하여 프롬프트의 인물 배제 조건을 심각하게 위반함." }, { "label": "C", "score": 5, "verdict_ko": "위를 올려다보는 시선과 자세는 좋으나, 캐릭터 기준 이미지의 회색 재킷을 누락하여 복장 오류가 있음." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S12sh9_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S12sh9" } }, "S13sh3::bgfirst_bg": { "input_fingerprint": "fd01ba1c0119298e", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 핸드폰을 귀에 댄 채 뒷좌석 쪽으로 상체를 반쯤 돌린 자세로 멈춰서 뒤를 주시하는 고산들의 상반신.\n\nLOCATION (lock): The driver’s seat of an older SUV traveling along a dark rural road at night, with the cabin dimly lit.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height inside the front passenger area, the pan finishes its rearward pivot and settles on 고산들 in a medium three-quarter view, his upper body twisted toward the back while the phone remains pressed to his ear. He occupies the left-center of frame, with the front-seat gap opening behind his gaze toward the unseen rear passenger; the brief stillness makes his divided attention legible.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: gap between the front seats (Unobstructed) — The gap faces the camera obliquely and opens toward the rear bench; used as The opening behind 고산들 extends his sightline toward the rear seat without revealing its occupant yet; driver's seat (Occupied) — Its side and upper back are visible from the front passenger area; used as Provides the immediate spatial boundary around his turned torso; mobile phone (Pressed to his ear) — Its screen face is turned away from the camera while its back and edge remain visible; used as Held beside his face as the active call prop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 핸드폰을 귀에 댄 채 뒷좌석 쪽으로 상체를 반쯤 돌린 자세로 멈춰서 뒤를 주시하는 고산들의 상반신.\n\nLOCATION (lock): The driver’s seat of an older SUV traveling along a dark rural road at night, with the cabin dimly lit.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From shoulder height inside the front passenger area, the pan finishes its rearward pivot and settles on 고산들 in a medium three-quarter view, his upper body twisted toward the back while the phone remains pressed to his ear. He occupies the left-center of frame, with the front-seat gap opening behind his gaze toward the unseen rear passenger; the brief stillness makes his divided attention legible.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: gap between the front seats (Unobstructed) — The gap faces the camera obliquely and opens toward the rear bench; used as The opening behind 고산들 extends his sightline toward the rear seat without revealing its occupant yet; driver's seat (Occupied) — Its side and upper back are visible from the front passenger area; used as Provides the immediate spatial boundary around his turned torso; mobile phone (Pressed to his ear) — Its screen face is turned away from the camera while its back and edge remain visible; used as Held beside his face as the active call prop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S13sh3__bgfirst_bg.png", "asset_id": "5abef8d1-552a-42df-9bf2-6281e1a13e2f", "input_asset_ids": [ "23e37876-8bf7-43b5-9ce7-b0c2e83c5f6c", "f16970f9-c021-454b-a773-e4692386ec50" ] }, "S13sh3": { "input_fingerprint": "fad777c03871c117", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 핸드폰을 귀에 댄 채 뒷좌석 쪽으로 상체를 반쯤 돌린 자세로 멈춰서 뒤를 주시하는 고산들의 상반신.\n\nLOCATION (lock): The driver’s seat of an older SUV traveling along a dark rural road at night, with the cabin dimly lit. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains bound hand and foot and lying unconscious across the rear seat, still wearing her bone necklace. Sandeul also retains the photograph frame taken from Minsook's house.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 핸드폰을 귀에 댄 채 뒷좌석 쪽으로 상체를 반쯤 돌린 자세로 멈춰서 뒤를 주시하는 고산들의 상반신.\n\nLOCATION (lock): The driver’s seat of an older SUV traveling along a dark rural road at night, with the cabin dimly lit. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains bound hand and foot and lying unconscious across the rear seat, still wearing her bone necklace. Sandeul also retains the photograph frame taken from Minsook's house.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 핸드폰을 귀에 댄 채 뒷좌석 쪽으로 상체를 반쯤 돌린 자세로 멈춰서 뒤를 주시하는 고산들의 상반신.\n\nLOCATION (lock): The driver’s seat of an older SUV traveling along a dark rural road at night, with the cabin dimly lit. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains bound hand and foot and lying unconscious across the rear seat, still wearing her bone necklace. Sandeul also retains the photograph frame taken from Minsook's house.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S13sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S13sh3.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L13B03.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 5, "verdict_ko": "상체를 돌려 뒷좌석을 주시하라는 자세 지시를 놓치고 정면을 향하고 있으나, 물리적으로 불가능한 렌더링 오류는 피했습니다." }, { "label": "A", "score": 3, "verdict_ko": "주인공의 자세는 프롬프트에 부합하나, 뒷좌석의 인물이 비현실적으로 작은 2D 일러스트로 렌더링되어 실사 및 인체 규정을 어긴 하드 위반입니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 5, "verdict_ko": "상체를 돌려 뒷좌석을 주시하라는 자세 지시를 놓치고 정면을 향하고 있으나, 물리적으로 불가능한 렌더링 오류는 피했습니다." }, { "label": "A", "score": 3, "verdict_ko": "주인공의 자세는 프롬프트에 부합하나, 뒷좌석의 인물이 비현실적으로 작은 2D 일러스트로 렌더링되어 실사 및 인체 규정을 어긴 하드 위반입니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "텍스트에 명시된 '상체를 돌려 뒤를 주시하는' 핵심 자세를 놓치고 정면을 향해 감점이나, 실사 렌더링과 소품 배치 규정을 안정적으로 준수함." }, { "label": "B", "score": 3, "verdict_ko": "뒷좌석의 인물이 2D 일러스트처럼 비현실적인 크기로 렌더링되어 콜라주 및 물리적 불가능에 해당하는 심각한 하드 위반." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "텍스트에 명시된 '상체를 돌려 뒤를 주시하는' 핵심 자세를 놓치고 정면을 향해 감점이나, 실사 렌더링과 소품 배치 규정을 안정적으로 준수함." }, { "label": "A", "score": 3, "verdict_ko": "뒷좌석의 인물이 2D 일러스트처럼 비현실적인 크기로 렌더링되어 콜라주 및 물리적 불가능에 해당하는 심각한 하드 위반." } ] }, "combined": { "totals": { "A": 6, "B": 11 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 11 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 5, "verdict_ko": "상체를 돌려 뒷좌석을 주시하라는 자세 지시를 놓치고 정면을 향하고 있으나, 물리적으로 불가능한 렌더링 오류는 피했습니다." }, { "label": "A", "score": 3, "verdict_ko": "주인공의 자세는 프롬프트에 부합하나, 뒷좌석의 인물이 비현실적으로 작은 2D 일러스트로 렌더링되어 실사 및 인체 규정을 어긴 하드 위반입니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L13B03.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S13sh3__bgfirst_bg.png", "bg_asset_id": "5abef8d1-552a-42df-9bf2-6281e1a13e2f", "bg_record_key": "S13sh3::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S13sh4": { "input_fingerprint": "27300db0f8432036", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 차량 뒷좌석에 손발이 묶인 채 죽은 듯 눈을 감고 누워 있는 김지후(강수리영)의 창백한 얼굴.\n\nLOCATION (lock): Across the rear bench of the moving older SUV, in the dim nighttime cabin. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same old SUV cabin, dark upholstery, rear-seat layout, and nighttime road light. Exclude the driver's phone-call pose and show only the bound woman lying across the back seat.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is stretched across the SUV's rear seat with her torso and head supported by the seat, both hands and both feet bound, and all four limbs held close to her body.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains stretched across the SUV's rear seat with her torso and head supported by the seat, both hands and both feet bound, and all four limbs held close to her body, unconscious, with her bone necklace still around her neck.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 4 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "필수 소품인 뼈 목걸이와 두 손발이 함께 묶인 결박 상태를 정확히 반영하여 지시문을 가장 잘 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "지시된 뼈 목걸이가 누락되었으며, 의상에 이전 샷에 없던 흰색 끈 장식이 추가되어 일관성을 떨어뜨립니다." }, { "label": "C", "score": 4, "verdict_ko": "손목에 밧줄이 감겨 있으나 두 손이 서로 묶이지 않고 떨어져 있어 결박 상태의 물리적 논리를 심각하게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S13sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S13sh3" } }, "S14sh2::bgfirst_bg": { "input_fingerprint": "d4317fe440f7498b", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 구급차 뒷문턱에 걸터앉은 채 한쪽 팔에 팔걸이 밴드를 매고 지친 표정으로 땅을 보는 정인우의 전신.\n\nLOCATION (lock): At the open rear threshold of an ambulance parked in the residential alley.\n\nTIME OF DAY (lock): evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 정인우's seated eye level, the forward dolly reaches its endpoint in a wide, nearly axial composition that preserves his entire seated figure at the ambulance threshold. He sits slightly off-center with his supported arm crossing his body and his lowered face separated from the police and residents active deeper in the alley.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: ambulance rear entrance (Open) — The open rear side faces the camera, with the threshold directly beneath him; used as Frames 정인우 from behind and establishes where he is receiving aid; police car light bars (Flashing red and blue); used as Creates intermittent colored illumination across the alley and the seated figure; multi-unit house (Police moving through the building) — Its alley-facing frontage is visible beyond the ambulance area; used as Provides the incident location behind the emergency response activity; neighborhood residents (Gathered and murmuring); used as Creates a dispersed background line of witnesses looking toward the residence.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 구급차 뒷문턱에 걸터앉은 채 한쪽 팔에 팔걸이 밴드를 매고 지친 표정으로 땅을 보는 정인우의 전신.\n\nLOCATION (lock): At the open rear threshold of an ambulance parked in the residential alley.\n\nTIME OF DAY (lock): evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 정인우's seated eye level, the forward dolly reaches its endpoint in a wide, nearly axial composition that preserves his entire seated figure at the ambulance threshold. He sits slightly off-center with his supported arm crossing his body and his lowered face separated from the police and residents active deeper in the alley.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: ambulance rear entrance (Open) — The open rear side faces the camera, with the threshold directly beneath him; used as Frames 정인우 from behind and establishes where he is receiving aid; police car light bars (Flashing red and blue); used as Creates intermittent colored illumination across the alley and the seated figure; multi-unit house (Police moving through the building) — Its alley-facing frontage is visible beyond the ambulance area; used as Provides the incident location behind the emergency response activity; neighborhood residents (Gathered and murmuring); used as Creates a dispersed background line of witnesses looking toward the residence.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S14sh2__bgfirst_bg.png", "asset_id": "b72e3401-2b0a-4aab-9759-d7c2be1eaa75", "input_asset_ids": [ "bb7ae2b7-0264-4ae2-a6d7-743ac9913e92", "65cb5ccb-2c65-4cc3-8e5e-c037b0aa24c3", "d5c560b6-50f1-45a8-aab7-8d195787f220" ] }, "S14sh2": { "input_fingerprint": "c1c5063c8068fdf1", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): evening.\n\nSHOT TEXT (authoritative, Korean): 구급차 뒷문턱에 걸터앉은 채 한쪽 팔에 팔걸이 밴드를 매고 지친 표정으로 땅을 보는 정인우의 전신.\n\nLOCATION (lock): At the open rear threshold of an ambulance parked in the residential alley. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo's injured arm remains supported in the sling.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): evening.\n\nSHOT TEXT (authoritative, Korean): 구급차 뒷문턱에 걸터앉은 채 한쪽 팔에 팔걸이 밴드를 매고 지친 표정으로 땅을 보는 정인우의 전신.\n\nLOCATION (lock): At the open rear threshold of an ambulance parked in the residential alley. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo's injured arm remains supported in the sling.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): evening.\n\nSHOT TEXT (authoritative, Korean): 구급차 뒷문턱에 걸터앉은 채 한쪽 팔에 팔걸이 밴드를 매고 지친 표정으로 땅을 보는 정인우의 전신.\n\nLOCATION (lock): At the open rear threshold of an ambulance parked in the residential alley. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo's injured arm remains supported in the sling.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S14sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S14sh2.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L11B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_multifamily_residence_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시문에 없는 인물을 배제하고 정인우의 전신, 팔걸이 밴드, 지친 자세, 그리고 레퍼런스의 의상을 충실히 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "지시문에 명시되지 않은 다수의 인물(경찰, 행인 등)을 임의로 배경에 추가하여 등장인물 제한 규정을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시문에 없는 인물을 배제하고 정인우의 전신, 팔걸이 밴드, 지친 자세, 그리고 레퍼런스의 의상을 충실히 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "지시문에 명시되지 않은 다수의 인물(경찰, 행인 등)을 임의로 배경에 추가하여 등장인물 제한 규정을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "주어진 샷 텍스트에 따라 정인우 단 한 명만 프레임에 담아냈으며, 지정된 의상과 슬링 착용, 시선 처리 모두 지시사항을 훌륭히 충족했습니다." }, { "label": "B", "score": 3, "verdict_ko": "지문에 언급되지 않은 구경꾼과 경찰 등 다수의 인물을 임의로 추가하여 'invented people'에 해당하는 심각한 규칙 위반을 범했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "주어진 샷 텍스트에 따라 정인우 단 한 명만 프레임에 담아냈으며, 지정된 의상과 슬링 착용, 시선 처리 모두 지시사항을 훌륭히 충족했습니다." }, { "label": "A", "score": 3, "verdict_ko": "지문에 언급되지 않은 구경꾼과 경찰 등 다수의 인물을 임의로 추가하여 'invented people'에 해당하는 심각한 규칙 위반을 범했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시문에 없는 인물을 배제하고 정인우의 전신, 팔걸이 밴드, 지친 자세, 그리고 레퍼런스의 의상을 충실히 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "지시문에 명시되지 않은 다수의 인물(경찰, 행인 등)을 임의로 배경에 추가하여 등장인물 제한 규정을 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L11B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_multifamily_residence_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S14sh2__bgfirst_bg.png", "bg_asset_id": "b72e3401-2b0a-4aab-9759-d7c2be1eaa75", "bg_record_key": "S14sh2::bgfirst_bg", "chain_winner": false, "authority": "plate", "seed_attached": true }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S14sh6": { "input_fingerprint": "887295d70a229405", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): evening.\n\nSHOT TEXT (authoritative, Korean): 다친 팔을 부여잡은 채 쓴웃음을 지으며 노 형사를 올려다보는 정인우의 허탈한 표정.\n\nLOCATION (lock): The emergency-response area in the residential alley below the multi-family home. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same ambulance threshold, flashing emergency lights, apartment exterior, and evening atmosphere. Exclude the wider crowd emphasis and retain the man's arm sling and seated position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the sling on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 5, "B": 2, "C": 3 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "레퍼런스의 카메라 구도를 그대로 복사하는 오류를 범했으나, 구급차 배경과 단독 인물 제약을 유일하게 준수함." }, { "label": "B", "score": 2, "verdict_ko": "구급차가 사라지고 길거리에 좌석만 놓인 물리적으로 불가능한 배경 구조(Hard Violation)를 생성함." }, { "label": "C", "score": 3, "verdict_ko": "전면에 주인공과 동일한 의상을 입은 복제된 인물(Hard Violation)을 배치하여 단독 인물 규정을 위반함." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S14sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S14sh2" } }, "S15sh4::bgfirst_bg": { "input_fingerprint": "b025b15745788371", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 눈을 번쩍 뜬 채 몸이 묶인 상태로 누워 두려움 가득한 눈빛으로 앞을 올려다보는 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): The rear seating area inside the parked older SUV before dawn, with faint dawn light entering through the windows.\n\nTIME OF DAY (lock): dawn.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the rear half of the SUV, the camera finishes slightly above 김지후(강수리영)'s eye line and angles down as the move tightens onto his suddenly opened eyes. His fearful face occupies most of the frame, but the line of his restrained shoulders and a small edge of the bindings remain visible below to establish that he is still lying immobilized.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) in the middle-center of the frame, foreground, looks toward 고산들 beyond the frame.\n- KEY BACKGROUND ELEMENTS: rear bench seat (Occupied) — The seat surface runs beneath his shoulders and recedes toward the side of the cabin; used as Supports the restrained figure while its seam supplies a restrained diagonal beneath his face; bindings (Securing his limbs); used as Makes the captive posture legible at the lower frame boundary; SUV side window (Closed) — The window remains at the background edge, with the pale dawn exterior only narrowly visible through it; used as Provides a narrow environmental edge behind his head rather than a separate focal point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 눈을 번쩍 뜬 채 몸이 묶인 상태로 누워 두려움 가득한 눈빛으로 앞을 올려다보는 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): The rear seating area inside the parked older SUV before dawn, with faint dawn light entering through the windows.\n\nTIME OF DAY (lock): dawn.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the rear half of the SUV, the camera finishes slightly above 김지후(강수리영)'s eye line and angles down as the move tightens onto his suddenly opened eyes. His fearful face occupies most of the frame, but the line of his restrained shoulders and a small edge of the bindings remain visible below to establish that he is still lying immobilized.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) in the middle-center of the frame, foreground, looks toward 고산들 beyond the frame.\n- KEY BACKGROUND ELEMENTS: rear bench seat (Occupied) — The seat surface runs beneath his shoulders and recedes toward the side of the cabin; used as Supports the restrained figure while its seam supplies a restrained diagonal beneath his face; bindings (Securing his limbs); used as Makes the captive posture legible at the lower frame boundary; SUV side window (Closed) — The window remains at the background edge, with the pale dawn exterior only narrowly visible through it; used as Provides a narrow environmental edge behind his head rather than a separate focal point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S15sh4__bgfirst_bg.png", "asset_id": "ce1f970b-9d8e-47fb-8c63-f907224b459d", "input_asset_ids": [ "358e8646-775a-42d9-b286-5ca954003fdc", "afffd034-74bb-495b-8670-6539c8f24382" ] }, "S15sh4": { "input_fingerprint": "fcad6d6206b95db2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 눈을 번쩍 뜬 채 몸이 묶인 상태로 누워 두려움 가득한 눈빛으로 앞을 올려다보는 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): The rear seating area inside the parked older SUV before dawn, with faint dawn light entering through the windows. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is stretched across the SUV's rear seat with her torso and head supported by the seat, both hands and both feet bound, and all four limbs held close to her body.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains stretched across the SUV's rear seat with her torso and head supported by the seat, both hands and both feet bound, and all four limbs held close to her body, with her bone necklace on, now awake.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 눈을 번쩍 뜬 채 몸이 묶인 상태로 누워 두려움 가득한 눈빛으로 앞을 올려다보는 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): The rear seating area inside the parked older SUV before dawn, with faint dawn light entering through the windows. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is stretched across the SUV's rear seat with her torso and head supported by the seat, both hands and both feet bound, and all four limbs held close to her body.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains stretched across the SUV's rear seat with her torso and head supported by the seat, both hands and both feet bound, and all four limbs held close to her body, with her bone necklace on, now awake.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 눈을 번쩍 뜬 채 몸이 묶인 상태로 누워 두려움 가득한 눈빛으로 앞을 올려다보는 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): The rear seating area inside the parked older SUV before dawn, with faint dawn light entering through the windows. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is stretched across the SUV's rear seat with her torso and head supported by the seat, both hands and both feet bound, and all four limbs held close to her body.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains stretched across the SUV's rear seat with her torso and head supported by the seat, both hands and both feet bound, and all four limbs held close to her body, with her bone necklace on, now awake.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S15sh4__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S15sh4.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L13B02.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 샷 크기(얼굴 중심)와 결박되어 누운 자세, 두려운 표정을 모두 정확하게 구현했습니다." }, { "label": "B", "score": 0, "verdict_ko": "신체 구조가 물리적으로 불가능하게 묘사되는 치명적인 오류(화면 우측에 추가적인 다리 등장)가 발생하여 규정을 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 샷 크기(얼굴 중심)와 결박되어 누운 자세, 두려운 표정을 모두 정확하게 구현했습니다." }, { "label": "B", "score": 0, "verdict_ko": "신체 구조가 물리적으로 불가능하게 묘사되는 치명적인 오류(화면 우측에 추가적인 다리 등장)가 발생하여 규정을 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 0, "verdict_ko": "다리와 하반신이 여러 개 생성된 심각한 신체 구조 오류(Hard Violation)가 발생하여 실격입니다." }, { "label": "B", "score": 9, "verdict_ko": "묶인 채 누워 두려움에 찬 눈으로 올려다보는 지후의 얼굴과 상반신을 프레임에 정확히 담아냈으며, 차량 내부 배경과 소품도 지시 사항을 훌륭하게 따랐습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 0, "verdict_ko": "다리와 하반신이 여러 개 생성된 심각한 신체 구조 오류(Hard Violation)가 발생하여 실격입니다." }, { "label": "A", "score": 9, "verdict_ko": "묶인 채 누워 두려움에 찬 눈으로 올려다보는 지후의 얼굴과 상반신을 프레임에 정확히 담아냈으며, 차량 내부 배경과 소품도 지시 사항을 훌륭하게 따랐습니다." } ] }, "combined": { "totals": { "A": 19, "B": 0 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 19, "B": 0 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 샷 크기(얼굴 중심)와 결박되어 누운 자세, 두려운 표정을 모두 정확하게 구현했습니다." }, { "label": "B", "score": 0, "verdict_ko": "신체 구조가 물리적으로 불가능하게 묘사되는 치명적인 오류(화면 우측에 추가적인 다리 등장)가 발생하여 규정을 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L13B02.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S15sh4__bgfirst_bg.png", "bg_asset_id": "ce1f970b-9d8e-47fb-8c63-f907224b459d", "bg_record_key": "S15sh4::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S15sh5": { "input_fingerprint": "709babe060e0df14", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 뒷좌석으로 고개를 돌린 채 말없이 김지후(강수리영)를 응시하는 고산들의 무거운 시선.\n\nLOCATION (lock): The front cabin of the older SUV near a harbor at dawn, faintly lit by the brightening sky outside. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains bound and lying in the rear seat behind Sandeul, still wearing her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 4, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "고산들이 지후를 응시하는 시선과 '짧게 깎은 머리' 지시를 정확히 구현했으며, 이전 샷의 지후 상태와 시선을 가장 잘 유지했습니다." }, { "label": "A", "score": 4, "verdict_ko": "고산들이 지후를 바라보는 행동은 맞으나 우핸들 차량으로 렌더링되었고, 지후가 눈을 감고 있어 이전 샷의 상태가 유지되지 않았습니다." }, { "label": "B", "score": 3, "verdict_ko": "고산들이 지후가 아닌 카메라를 응시해 샷 텍스트의 핵심을 놓쳤으며, 헤어스타일 지시를 무시하고 모자를 착용했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S15sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S15sh4" } }, "S16sh1::bgfirst_bg": { "input_fingerprint": "005514ab68de8066", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 경찰서 형사과 안, 컴퓨터 모니터 화면에 바짝 얼굴을 들이댄 정인우의 미간이 잔뜩 찌푸려진 굳은 얼굴.\n\nLOCATION (lock): A workstation inside a police detective office, directly before a computer monitor under office lighting.\n\nTIME OF DAY (lock): dawn.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level beside the computer monitor, the camera presses into a close, slightly frontal three-quarter view of 정인우 leaning toward the screen. His furrowed brow and rigid face occupy most of the frame, while only an oblique sliver of the active monitor remains along the near-left edge as the source of his hostile concentration.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: computer monitor (Displaying CCTV footage) — Only a narrow oblique portion of the display face is visible, carrying the CCTV footage while most of the image remains cropped; used as Acts as the near-edge spatial reference and the object of 정인우's fixed attention; arm support band (Supporting one arm); used as Remains visible on his injured side and links this shot to the prior alley scene; detective desk (In use) — Its near edge recedes out of frame beneath the monitor; used as Supplies a minimal office boundary behind the tight portrait.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 경찰서 형사과 안, 컴퓨터 모니터 화면에 바짝 얼굴을 들이댄 정인우의 미간이 잔뜩 찌푸려진 굳은 얼굴.\n\nLOCATION (lock): A workstation inside a police detective office, directly before a computer monitor under office lighting.\n\nTIME OF DAY (lock): dawn.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level beside the computer monitor, the camera presses into a close, slightly frontal three-quarter view of 정인우 leaning toward the screen. His furrowed brow and rigid face occupy most of the frame, while only an oblique sliver of the active monitor remains along the near-left edge as the source of his hostile concentration.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: computer monitor (Displaying CCTV footage) — Only a narrow oblique portion of the display face is visible, carrying the CCTV footage while most of the image remains cropped; used as Acts as the near-edge spatial reference and the object of 정인우's fixed attention; arm support band (Supporting one arm); used as Remains visible on his injured side and links this shot to the prior alley scene; detective desk (In use) — Its near edge recedes out of frame beneath the monitor; used as Supplies a minimal office boundary behind the tight portrait.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S16sh1__bgfirst_bg.png", "asset_id": "a518fd77-cf43-45e5-8055-963af1ae604b", "input_asset_ids": [ "cdb65e79-453a-4c9a-9f78-55046564e663", "70176d16-c940-4cb7-a497-3b95baa77114" ] }, "S16sh1": { "input_fingerprint": "e17f40e87b98d94f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 경찰서 형사과 안, 컴퓨터 모니터 화면에 바짝 얼굴을 들이댄 정인우의 미간이 잔뜩 찌푸려진 굳은 얼굴.\n\nLOCATION (lock): A workstation inside a police detective office, directly before a computer monitor under office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the sling on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 경찰서 형사과 안, 컴퓨터 모니터 화면에 바짝 얼굴을 들이댄 정인우의 미간이 잔뜩 찌푸려진 굳은 얼굴.\n\nLOCATION (lock): A workstation inside a police detective office, directly before a computer monitor under office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the sling on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 경찰서 형사과 안, 컴퓨터 모니터 화면에 바짝 얼굴을 들이댄 정인우의 미간이 잔뜩 찌푸려진 굳은 얼굴.\n\nLOCATION (lock): A workstation inside a police detective office, directly before a computer monitor under office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the sling on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S16sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S16sh1.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L14B01.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 '모니터에 바짝 얼굴을 들이댄' 자세와 잔뜩 찌푸린 미간의 표정을 카메라 구도 내에서 가장 정확하게 연출했습니다." }, { "label": "B", "score": 5, "verdict_ko": "새벽이라는 시간대 묘사는 우수하나, 화면에 바짝 다가간 역동적인 밀착 자세가 부족해 샷 텍스트의 핵심 행동을 충분히 살리지 못했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 '모니터에 바짝 얼굴을 들이댄' 자세와 잔뜩 찌푸린 미간의 표정을 카메라 구도 내에서 가장 정확하게 연출했습니다." }, { "label": "B", "score": 5, "verdict_ko": "새벽이라는 시간대 묘사는 우수하나, 화면에 바짝 다가간 역동적인 밀착 자세가 부족해 샷 텍스트의 핵심 행동을 충분히 살리지 못했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "모니터에 바짝 다가간 얼굴과 찌푸린 미간이라는 숏 텍스트의 핵심 묘사를 프레이밍과 피사체 간 거리감을 통해 가장 정확하게 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "지정된 인상과 소품(팔걸이)은 잘 표현되었으나, 모니터에 '바짝' 얼굴을 들이밀었다는 지문에 비해 피사체와 화면 간의 거리가 멀어 긴장감이 떨어집니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "모니터에 바짝 다가간 얼굴과 찌푸린 미간이라는 숏 텍스트의 핵심 묘사를 프레이밍과 피사체 간 거리감을 통해 가장 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "지정된 인상과 소품(팔걸이)은 잘 표현되었으나, 모니터에 '바짝' 얼굴을 들이밀었다는 지문에 비해 피사체와 화면 간의 거리가 멀어 긴장감이 떨어집니다." } ] }, "combined": { "totals": { "A": 15, "B": 11 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 15, "B": 11 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 '모니터에 바짝 얼굴을 들이댄' 자세와 잔뜩 찌푸린 미간의 표정을 카메라 구도 내에서 가장 정확하게 연출했습니다." }, { "label": "B", "score": 5, "verdict_ko": "새벽이라는 시간대 묘사는 우수하나, 화면에 바짝 다가간 역동적인 밀착 자세가 부족해 샷 텍스트의 핵심 행동을 충분히 살리지 못했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L14B01.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S16sh1__bgfirst_bg.png", "bg_asset_id": "a518fd77-cf43-45e5-8055-963af1ae604b", "bg_record_key": "S16sh1::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S16sh7": { "input_fingerprint": "2a760f38e63e32b7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 상황을 묻듯 고개를 내민 노 형사의 옆에서 넋이 나간 표정으로 허공을 멍하니 응시하는 정인우의 상체.\n\nLOCATION (lock): Beside a detective-office desk and computer station in the early-morning interior. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo still wears the sling on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 노 형사 (40대 한국인 남성, 평범한 중년 얼굴, 짧고 단정한 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 4, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "이전 샷의 모니터(CCTV 화면)와 책상 위 소품 배치를 완벽히 유지하며, 넋이 나간 정인우와 고개를 내민 노 형사의 상체를 정확하게 연출함." }, { "label": "A", "score": 4, "verdict_ko": "인물들의 표정과 행동은 무난하나, 이전 샷에서 인물 앞에 있던 모니터가 인물 뒤로 이동하는 등 공간과 소품의 연속성이 완전히 깨짐." }, { "label": "B", "score": 4, "verdict_ko": "지시된 앵글과 인물의 자세는 반영되었으나, A와 마찬가지로 확립된 데스크 공간의 구조와 소품 배치를 무시하여 연속성 유지에 실패함." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S16sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 노 형사: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S16sh1" } }, "S17sh3::bgfirst_bg": { "input_fingerprint": "eda5f2620e6964c1", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 뱃머리 끝에 홀로 꼿꼿하게 선 채 탁 트인 바다를 굽어보는 고산들의 단단한 뒷모습.\n\nLOCATION (lock): The exposed bow of a moving fishing boat on open sea in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height on the bow centerline several steps behind 고산들, a static wide frame places his rigid back in the middle against the open sea. His braced figure remains relatively small within the environment, with the bow converging below him and uninterrupted water beyond, turning his isolation into the dominant visual fact.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 고산들 in the middle-center of the frame, midground; open sea in the upper-center of the frame, background; boat bow in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: boat bow (Advancing with the fishing boat) — Its forward-pointing sides converge toward the center beneath 고산들; used as Forms the converging foreground path that directs the eye into 고산들's back; open sea (Blue and rolling); used as Creates the open background against which 고산들 is isolated; central passage (Clear) — Its long axis aligns with the camera and points from the bow toward the stern; used as Extends behind the camera path and establishes the route of the following backward move.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 뱃머리 끝에 홀로 꼿꼿하게 선 채 탁 트인 바다를 굽어보는 고산들의 단단한 뒷모습.\n\nLOCATION (lock): The exposed bow of a moving fishing boat on open sea in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height on the bow centerline several steps behind 고산들, a static wide frame places his rigid back in the middle against the open sea. His braced figure remains relatively small within the environment, with the bow converging below him and uninterrupted water beyond, turning his isolation into the dominant visual fact.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 고산들 in the middle-center of the frame, midground; open sea in the upper-center of the frame, background; boat bow in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: boat bow (Advancing with the fishing boat) — Its forward-pointing sides converge toward the center beneath 고산들; used as Forms the converging foreground path that directs the eye into 고산들's back; open sea (Blue and rolling); used as Creates the open background against which 고산들 is isolated; central passage (Clear) — Its long axis aligns with the camera and points from the bow toward the stern; used as Extends behind the camera path and establishes the route of the following backward move.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S17sh3__bgfirst_bg.png", "asset_id": "8987aa7b-a136-4d8c-93ff-a016dd8dfaa8", "input_asset_ids": [ "31406af8-1e2f-4396-9c0b-dd6dec8514dd", "08723220-59c5-4941-bd8d-65699b31da46" ] }, "S17sh3": { "input_fingerprint": "3c4ee5dc4ff1719a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 뱃머리 끝에 홀로 꼿꼿하게 선 채 탁 트인 바다를 굽어보는 고산들의 단단한 뒷모습.\n\nLOCATION (lock): The exposed bow of a moving fishing boat on open sea in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 뱃머리 끝에 홀로 꼿꼿하게 선 채 탁 트인 바다를 굽어보는 고산들의 단단한 뒷모습.\n\nLOCATION (lock): The exposed bow of a moving fishing boat on open sea in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 뱃머리 끝에 홀로 꼿꼿하게 선 채 탁 트인 바다를 굽어보는 고산들의 단단한 뒷모습.\n\nLOCATION (lock): The exposed bow of a moving fishing boat on open sea in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S17sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S17sh3.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L16B05.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 로케이션과 인물의 뒷모습을 잘 구현했으며, 레퍼런스 이미지의 의상(가죽 재킷, 카고 팬츠)을 비교적 충실히 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "인물의 구도와 자세는 좋으나, 레퍼런스와 전혀 다른 파란색 후드 재킷을 입고 있어 의상 일관성에서 감점되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 로케이션과 인물의 뒷모습을 잘 구현했으며, 레퍼런스 이미지의 의상(가죽 재킷, 카고 팬츠)을 비교적 충실히 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "인물의 구도와 자세는 좋으나, 레퍼런스와 전혀 다른 파란색 후드 재킷을 입고 있어 의상 일관성에서 감점되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "레퍼런스의 의상(갈색 가죽 재킷과 카고 팬츠)을 정확히 반영하여 단단한 뒷모습을 훌륭하게 구현했으며, 뱃머리 끝에 서 있는 위치 설정도 지문에 잘 부합합니다." }, { "label": "A", "score": 5, "verdict_ko": "뒷모습은 잘 표현되었으나 뱃머리 끝이 아닌 선실 바로 옆에 위치해 있으며, 레퍼런스와 완전히 다른 파란색 후드 재킷을 입고 있어 캐릭터 일치도에서 크게 감점됩니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "레퍼런스의 의상(갈색 가죽 재킷과 카고 팬츠)을 정확히 반영하여 단단한 뒷모습을 훌륭하게 구현했으며, 뱃머리 끝에 서 있는 위치 설정도 지문에 잘 부합합니다." }, { "label": "B", "score": 5, "verdict_ko": "뒷모습은 잘 표현되었으나 뱃머리 끝이 아닌 선실 바로 옆에 위치해 있으며, 레퍼런스와 완전히 다른 파란색 후드 재킷을 입고 있어 캐릭터 일치도에서 크게 감점됩니다." } ] }, "combined": { "totals": { "A": 18, "B": 11 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 18, "B": 11 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 로케이션과 인물의 뒷모습을 잘 구현했으며, 레퍼런스 이미지의 의상(가죽 재킷, 카고 팬츠)을 비교적 충실히 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "인물의 구도와 자세는 좋으나, 레퍼런스와 전혀 다른 파란색 후드 재킷을 입고 있어 의상 일관성에서 감점되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L16B05.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S17sh3__bgfirst_bg.png", "bg_asset_id": "8987aa7b-a136-4d8c-93ff-a016dd8dfaa8", "bg_record_key": "S17sh3::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S17sh6": { "input_fingerprint": "a796be4a730654af", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 뱃머리에 선 고산들의 뒷모습을 향해 고개를 돌린 채 굳은 표정으로 쏘아보는 김지후(강수리영)의 날 선 시선.\n\nLOCATION (lock): A passenger area on the aft deck of the fishing boat, looking forward along the exterior passage to the bow. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same fishing-boat deck, bow geometry, open blue sea, and bright daylight. Exclude the stern anglers from prominence and preserve the man standing at the bow while the woman watches him.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped bone necklace aboard the boat.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 4, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "뱃머리의 남자를 향해 고개를 돌려 쏘아보는 시선을 가장 잘 구현했으나, 기준 이미지에 있던 재킷이 누락되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 시선이 남자를 향하지 않고 다른 곳을 보고 있어 프롬프트의 핵심 연출 지시를 어겼습니다." }, { "label": "C", "score": 3, "verdict_ko": "화면 우측 하단에 촬영용 카메라 삼각대가 노출되는 치명적인 객체 생성 오류가 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S17sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S17sh3" }, "lane_policy": "ab_select_bypass:prev" }, "S18sh1": { "input_fingerprint": "ba8e5eccc2ca7a4b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 어두운 구형 산타페 조수석에 앉아 양손으로 낡은 사진 액자를 꽉 쥔 채 뚫어지게 내려다보는 김지후(강수리영)의 상체.\n\nLOCATION (lock): The front passenger seat of an older SUV before dawn, dimly lit by the pale sky through the windows. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu is no longer bound, still wears her bone necklace, and holds the same photograph frame Sandeul removed from Minsook's home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 5, "B": 4, "C": 7 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "액자 속 사진은 일치하나 차량 내부 구조가 왜곡되었고, 소품의 정면이 카메라를 향해 행동이 부자연스럽습니다." }, { "label": "B", "score": 4, "verdict_ko": "조수석 위치는 정확하지만, 액자 속 인물이 레퍼런스와 다르고 완전히 카메라를 향해 들고 있어 지시와 어긋납니다." }, { "label": "C", "score": 7, "verdict_ko": "조수석이 아닌 운전석에 앉은 위치 오류가 있으나, 소품의 방향과 사진을 응시하는 행동 지시를 유일하게 완벽히 구현했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S15sh5_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 낡은 사진 액자: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S15sh5" } }, "S18sh5": { "input_fingerprint": "fa60c4cef14d29b0", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 강렬한 시선을 피하지 않고 무표정하게 마주 보는 고산들의 덤덤한 얼굴.\n\nLOCATION (lock): The front cabin of the older SUV, between the driver’s and passenger’s seats in faint dawn light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same old SUV dashboard, passenger-seat area, dim dawn light, and the worn photo frame. Exclude the woman's downward-looking pose and frame the driver's impassive response.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The photograph frame remains in Jihu's hands, and her bone necklace remains around her neck.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 6, "B": 3, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지정된 실내 카메라 시점과 두 인물의 시선 교환을 정확히 구현했으나, 이전 샷과 좌석 배치가 역전되고 소품 방향이 어색함." }, { "label": "B", "score": 3, "verdict_ko": "카메라가 차량 외부로 벗어났으며, 여성 캐릭터가 누락되어 핵심 상호작용이 전혀 표현되지 않음." }, { "label": "C", "score": 4, "verdict_ko": "이전 샷의 좌석 배치는 유지했으나 카메라가 차량 외부(보닛)로 벗어났고, 남자가 시선을 피하고 있어 지시문과 어긋남." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S18sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S18sh1" } }, "S19sh3": { "input_fingerprint": "825fa89c8718f3d9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 주변을 경계하는 눈빛을 띤 채 김지후(강수리영)의 귓가로 고개를 바짝 들이민 김여상의 비밀스러운 옆모습.\n\nLOCATION (lock): A seated passenger spot on the fishing boat’s open deck, slightly apart from the other anglers. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same boat deck, railings, open sea, and clear daylight. Exclude the solitary bow composition and place the cautious fisherman beside the seated woman.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 김여상 (40대 중반 한국인 남성, 평범한 중년 얼굴, 단정한 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 4, "C": 2 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "이전 샷에서 확립된 남성의 의상(회색 재킷)을 완벽하게 유지했으며, 귓가로 다가간 프로필과 경계하는 시선을 지시문대로 정확히 연출했습니다." }, { "label": "A", "score": 4, "verdict_ko": "인물들의 동작과 표정은 좋으나, 이전 샷과 동일하게 유지되어야 할 남성의 의상을 무시하고 붉은색 옷을 입힌 점이 큰 감점 요인입니다." }, { "label": "C", "score": 2, "verdict_ko": "이전 샷의 의상 유지를 위반했을 뿐만 아니라, 텍스트에 명시되지 않은 제3의 인물들을 배경에 추가하여 치명적인 위반(Hard Violation)을 범했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S17sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김여상: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S17sh6" } }, "S19sh8": { "input_fingerprint": "3d8cc23b68d4a6fa", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 위에서 내려다보는 고산들을 향해 턱을 치켜든 채 한 치의 흔들림 없는 결연한 눈빛을 발산하는 김지후(강수리영)의 표정.\n\nLOCATION (lock): The open deck passage near the bow of the moving fishing boat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same fishing-boat deck, bow and stern layout, sea color, and daylight. Exclude the earlier chatting anglers from the close frame and show the woman looking firmly up at the standing man.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "숏 텍스트의 지시(고산들을 향해)와 단일 인물 제약을 완벽히 준수하며 이전 샷의 선실 배경과 소품을 정확히 유지했습니다." }, { "label": "B", "score": 5, "verdict_ko": "인물 규칙은 지켰으나 배경이 선실이 아닌 선수 뷰로 바뀌어 인물의 위치 및 장소 연속성을 훼손했습니다." }, { "label": "A", "score": 3, "verdict_ko": "숏 텍스트에 명시되지 않은 타인의 신체 일부(남성의 팔과 몸통)를 프레임에 포함시켜 인물 제약 규칙을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S19sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S19sh3" } }, "S19sh11": { "input_fingerprint": "bb0b400a4cc2eaa5", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 끝없이 펼쳐진 푸른 바다 한가운데 홀로 우뚝 솟아 햇살을 받아 반짝이는 고립된 섬 금월도의 전경.\n\nLOCATION (lock): The isolated island seen across sunlit open water from the approaching fishing boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "A": 7, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시문이 요구한 '햇살을 받아 반짝이는 섬의 전경'을 정확한 구도로 구현했으며, 레퍼런스의 카메라 구도를 복사하지 말라는 지침을 잘 따랐습니다." }, { "label": "A", "score": 7, "verdict_ko": "섬의 전경과 햇살이 반짝이는 바다를 잘 표현했으나, 화면 상하단에 레터박스가 포함되어 아쉽습니다." }, { "label": "C", "score": 4, "verdict_ko": "레퍼런스 이미지의 카메라 구도를 그대로 복사하지 말라는 지침을 어겼으며, 배가 화면을 크게 차지해 '섬의 전경'이라는 묘사에 부합하지 않습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L16B01.png" } ], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S20sh2::bgfirst_bg": { "input_fingerprint": "aa0829724b93cf78", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어선 선착장에 서서 낯선 한국인 마을 사람들의 시선을 받으며 눈을 둥글게 뜬 채 당황한 표정을 짓는 김지후(강수리영)의 상체.\n\nLOCATION (lock): The landing area of a small island pier beside the newly arrived fishing boat.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 김지후(강수리영)'s chest height on the dock, end the inward dolly in a medium view through the surrounding villagers, using partial shoulders and heads at both edges as a compressed human frame. Place Jihoo near center with her body momentarily stalled and her rounded eyes moving past the camera-side crowd, while the gathered villagers remain contextual rather than individually prominent.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: gathered Korean villagers (Crowding toward the arrived boat, applauding and cheering for Jihoo); used as Partial foreground bodies create a corridor of attention converging on Jihoo; fishing boat and dock (The fishing boat has arrived at the dock) — The arrived boat sits behind the crowd, while the dock runs through Jihoo's position toward the village; used as Visible behind the gathering to establish the point of arrival; small island village (Fewer than roughly fifty households, surrounded by trees and forest in peak autumn foliage) — The colorful-roofed houses face the dock from beyond the gathered villagers; used as Distant environmental context beyond the crowd.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어선 선착장에 서서 낯선 한국인 마을 사람들의 시선을 받으며 눈을 둥글게 뜬 채 당황한 표정을 짓는 김지후(강수리영)의 상체.\n\nLOCATION (lock): The landing area of a small island pier beside the newly arrived fishing boat.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 김지후(강수리영)'s chest height on the dock, end the inward dolly in a medium view through the surrounding villagers, using partial shoulders and heads at both edges as a compressed human frame. Place Jihoo near center with her body momentarily stalled and her rounded eyes moving past the camera-side crowd, while the gathered villagers remain contextual rather than individually prominent.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: gathered Korean villagers (Crowding toward the arrived boat, applauding and cheering for Jihoo); used as Partial foreground bodies create a corridor of attention converging on Jihoo; fishing boat and dock (The fishing boat has arrived at the dock) — The arrived boat sits behind the crowd, while the dock runs through Jihoo's position toward the village; used as Visible behind the gathering to establish the point of arrival; small island village (Fewer than roughly fifty households, surrounded by trees and forest in peak autumn foliage) — The colorful-roofed houses face the dock from beyond the gathered villagers; used as Distant environmental context beyond the crowd.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh2__bgfirst_bg.png", "asset_id": "0624e70f-4939-4860-bdc5-9da1a09e7608", "input_asset_ids": [ "792c1899-121c-4752-b29e-d5e2f465717c", "338f96b2-4333-4b47-a7a4-51ae2225f29e" ] }, "S20sh2": { "input_fingerprint": "da90d93192df7ab6", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어선 선착장에 서서 낯선 한국인 마을 사람들의 시선을 받으며 눈을 둥글게 뜬 채 당황한 표정을 짓는 김지후(강수리영)의 상체.\n\nLOCATION (lock): The landing area of a small island pier beside the newly arrived fishing boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu arrives on Geumwoldo still wearing her arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어선 선착장에 서서 낯선 한국인 마을 사람들의 시선을 받으며 눈을 둥글게 뜬 채 당황한 표정을 짓는 김지후(강수리영)의 상체.\n\nLOCATION (lock): The landing area of a small island pier beside the newly arrived fishing boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu arrives on Geumwoldo still wearing her arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어선 선착장에 서서 낯선 한국인 마을 사람들의 시선을 받으며 눈을 둥글게 뜬 채 당황한 표정을 짓는 김지후(강수리영)의 상체.\n\nLOCATION (lock): The landing area of a small island pier beside the newly arrived fishing boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu arrives on Geumwoldo still wearing her arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S20sh2.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 상체 샷과 당황한 표정을 잘 표현했으며, 마을 사람들의 시선과 배경 묘사가 요구사항에 부합합니다." }, { "label": "B", "score": 3, "verdict_ko": "주인공의 표정과 구도는 좋으나, 배경 왼쪽 남성의 다리 부근에 기괴하게 일그러진 정체불명의 인물(머리)이 생성된 치명적인 오류(Hard Violation)가 있습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 상체 샷과 당황한 표정을 잘 표현했으며, 마을 사람들의 시선과 배경 묘사가 요구사항에 부합합니다." }, { "label": "B", "score": 3, "verdict_ko": "주인공의 표정과 구도는 좋으나, 배경 왼쪽 남성의 다리 부근에 기괴하게 일그러진 정체불명의 인물(머리)이 생성된 치명적인 오류(Hard Violation)가 있습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "마을 사람들의 낯선 시선을 받는 주인공의 당황한 표정과 상체 프레이밍을 잘 구현했으나, 배경 인물 중 일부의 형태가 다소 뭉개져 아쉽습니다." }, { "label": "B", "score": 4, "verdict_ko": "지시된 표정과 구도를 따랐으나 프롬프트에 없는 가방을 추가했고, 오른쪽 분홍색 옷을 입은 배경 인물의 박수 치는 손이 해부학적으로 불가능하게 묘사되어 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "마을 사람들의 낯선 시선을 받는 주인공의 당황한 표정과 상체 프레이밍을 잘 구현했으나, 배경 인물 중 일부의 형태가 다소 뭉개져 아쉽습니다." }, { "label": "A", "score": 4, "verdict_ko": "지시된 표정과 구도를 따랐으나 프롬프트에 없는 가방을 추가했고, 오른쪽 분홍색 옷을 입은 배경 인물의 박수 치는 손이 해부학적으로 불가능하게 묘사되어 감점되었습니다." } ] }, "combined": { "totals": { "A": 11, "B": 10 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 10 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 상체 샷과 당황한 표정을 잘 표현했으며, 마을 사람들의 시선과 배경 묘사가 요구사항에 부합합니다." }, { "label": "B", "score": 3, "verdict_ko": "주인공의 표정과 구도는 좋으나, 배경 왼쪽 남성의 다리 부근에 기괴하게 일그러진 정체불명의 인물(머리)이 생성된 치명적인 오류(Hard Violation)가 있습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh2__bgfirst_bg.png", "bg_asset_id": "0624e70f-4939-4860-bdc5-9da1a09e7608", "bg_record_key": "S20sh2::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S20sh6::bgfirst_bg": { "input_fingerprint": "8bef129717433676", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 김지후(강수리영)의 시선 끝에 온화한 미소를 띠고 선 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The far end of the island pier, facing the disembarkation point and village approach.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From 김지후(강수리영)'s approaching eye line, hold a static wide subjective view with 강영란(흰 머리 여자)'s full figure centered at the end of the dock and clear space separating her from the crowd. Her direct smile meets the lens as Jihoo would see it, while the dock's receding line makes the apparently warm welcome feel ceremonially arranged.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: end of the dock (Occupied by Youngran awaiting Jihoo) — The dock points from the camera position toward Youngran's centered full figure; used as Its receding line leads Jihoo's subjective gaze directly to Youngran; island village houses and trees (Trees and forest are in peak autumn foliage) — The colorful-roofed houses and surrounding autumn trees rise beyond the dock; used as Provides the picturesque village context beyond the isolated greeting figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe main pier extends seaward from the village-side approach and terminates at an outer pier end surrounded by the adjacent sea. Its main deck provides a continuous route between the village and the outer end, while a vessel berth edge directly contacts the water along the boat-access side. At the landward end, the approach continues into the small island village. The fishing-gear warehouse and Banwol Restaurant face this pier-to-village route, with the warehouse doorway opening onto the landing-side escape path and the restaurant entrance addressing the same general frontage. Small village houses occupy the inland area around the approach, with woodland interwoven among and behind them. This arrangement allows views inward from the outer pier toward the village, outward from the village frontage toward the berth and sea, and lateral coverage between the warehouse, restaurant frontage, and main pier deck.\n- pier end: seaward terminal end of the island pier\n- pier: main island pier deck\n- berth: water-contacting vessel berth edge\n- water: sea adjoining the island pier\n- approach: village-to-pier approach\n- houses: small island village houses around the pier approach\n- woodland: woodland surrounding the small island village\n- warehouse: fishing-gear warehouse beside the pier landing\n- door: pier-facing fishing-gear warehouse doorway\n- restaurant: Banwol Restaurant at the village-side pier frontage\n- entrance: pier-visible entrance of Banwol Restaurant\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 김지후(강수리영)의 시선 끝에 온화한 미소를 띠고 선 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The far end of the island pier, facing the disembarkation point and village approach.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From 김지후(강수리영)'s approaching eye line, hold a static wide subjective view with 강영란(흰 머리 여자)'s full figure centered at the end of the dock and clear space separating her from the crowd. Her direct smile meets the lens as Jihoo would see it, while the dock's receding line makes the apparently warm welcome feel ceremonially arranged.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: end of the dock (Occupied by Youngran awaiting Jihoo) — The dock points from the camera position toward Youngran's centered full figure; used as Its receding line leads Jihoo's subjective gaze directly to Youngran; island village houses and trees (Trees and forest are in peak autumn foliage) — The colorful-roofed houses and surrounding autumn trees rise beyond the dock; used as Provides the picturesque village context beyond the isolated greeting figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe main pier extends seaward from the village-side approach and terminates at an outer pier end surrounded by the adjacent sea. Its main deck provides a continuous route between the village and the outer end, while a vessel berth edge directly contacts the water along the boat-access side. At the landward end, the approach continues into the small island village. The fishing-gear warehouse and Banwol Restaurant face this pier-to-village route, with the warehouse doorway opening onto the landing-side escape path and the restaurant entrance addressing the same general frontage. Small village houses occupy the inland area around the approach, with woodland interwoven among and behind them. This arrangement allows views inward from the outer pier toward the village, outward from the village frontage toward the berth and sea, and lateral coverage between the warehouse, restaurant frontage, and main pier deck.\n- pier end: seaward terminal end of the island pier\n- pier: main island pier deck\n- berth: water-contacting vessel berth edge\n- water: sea adjoining the island pier\n- approach: village-to-pier approach\n- houses: small island village houses around the pier approach\n- woodland: woodland surrounding the small island village\n- warehouse: fishing-gear warehouse beside the pier landing\n- door: pier-facing fishing-gear warehouse doorway\n- restaurant: Banwol Restaurant at the village-side pier frontage\n- entrance: pier-visible entrance of Banwol Restaurant\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh6__bgfirst_bg.png", "asset_id": "299967af-30a3-457d-b1fb-ae1c8dca6cd8", "input_asset_ids": [ "3493f3ca-8555-48c0-b786-6d73b9037099" ] }, "S20sh6": { "input_fingerprint": "78690f9e372a2a4f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 시선 끝에 온화한 미소를 띠고 선 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The far end of the island pier, facing the disembarkation point and village approach. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same small-island dock, gathered villagers, colorful village backdrop, and daylight. Exclude the woman's isolated reaction and present the elderly woman waiting at the end of the dock.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear the arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 시선 끝에 온화한 미소를 띠고 선 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The far end of the island pier, facing the disembarkation point and village approach. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear the arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 시선 끝에 온화한 미소를 띠고 선 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The far end of the island pier, facing the disembarkation point and village approach. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear the arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 시선 끝에 온화한 미소를 띠고 선 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The far end of the island pier, facing the disembarkation point and village approach. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear the arrowhead-shaped animal-bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh6__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh6__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh6__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "C": 7, "B": 5, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "이전 샷의 걷는 포즈(오른발 들림)를 가장 정확히 유지했으나, 영란의 머리수건 색상이 레퍼런스와 다릅니다." }, { "label": "B", "score": 5, "verdict_ko": "영란의 묘사는 좋으나, 지후가 걷는 포즈를 유지하지 않고 평발로 서 있어 포즈 지침을 위반했습니다." }, { "label": "A", "score": 3, "verdict_ko": "영란의 의상이 지정된 레퍼런스(녹색 한복)와 완전히 다르며, 지후의 발 위치도 반대입니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh6__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh6__bgfirst_bg.png", "bg_asset_id": "299967af-30a3-457d-b1fb-ae1c8dca6cd8", "bg_record_key": "S20sh6::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S20sh2" } }, "S20sh9": { "input_fingerprint": "48284b478ba6a3a2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 강영란(흰 머리 여자)의 품에 꽉 안긴 채 두 팔을 갈 곳 잃고 허공에 띄운 김지후(강수리영)의 당황한 옆모습.\n\nLOCATION (lock): The open pier walkway among gathered island villagers in daylight. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dock-end position, village crowd, bright daylight, and welcoming atmosphere. Exclude the elderly woman's open-arm pose and show her already embracing the younger woman.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains in Youngran's embrace with the arrowhead-shaped animal-bone necklace around her neck.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 3, "B": 3, "C": 7 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "배경 좌측에 주인공이 중복 등장하는 치명적 오류가 있으며, 두 팔을 허공에 띄운 연출도 부족합니다." }, { "label": "B", "score": 3, "verdict_ko": "허공에 띄운 팔의 묘사는 훌륭하나, 배경 우측에 주인공이 똑같이 중복 등장하는 치명적인 오류가 있습니다." }, { "label": "C", "score": 7, "verdict_ko": "당황하여 두 팔을 갈 곳 잃고 허공에 띄운 옆모습을 가장 정확하고 자연스럽게 연출하여 지시문과 부합합니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S20sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S20sh6" } }, "S21sh11": { "input_fingerprint": "6d7dacf786eab0fd", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset transitioning to night.\n\nSHOT TEXT (authoritative, Korean): 핏기가 선명한 붉은 생고기 한 상이 테이블 위에 푸짐하게 차려진 순간의 조감도.\n\nLOCATION (lock): A dining table in the combined restaurant and general-store hall, beside the open kitchen in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The abundant spread of raw meat remains arranged on the table, and Jihu continues wearing her bone necklace.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 4, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요청된 '조감도'에 맞게 테이블을 위에서 내려다보는 구도를 완벽히 구현했으며, 노을빛 조명과 생고기의 질감, 레퍼런스의 테이블 형태를 모두 충실히 반영했습니다." }, { "label": "B", "score": 4, "verdict_ko": "높은 각도의 구도를 취했으나, 여러 개의 테이블이 비정상적으로 하나로 길게 합쳐지는 구조적 왜곡이 발생했습니다." }, { "label": "C", "score": 3, "verdict_ko": "장소의 기본 구조(주방, 선반, 창문의 위치)를 레퍼런스와 전혀 다르게 임의로 재배치하여 로케이션 고정 조건을 심각하게 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B02.png" } ], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S21sh16": { "input_fingerprint": "61d3df0906bf510e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset transitioning to night.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 목에 걸린 화살촉 모양의 뼈 목걸이를 뚫어지게 응시하는 강영란(흰 머리 여자)의 예리한 시선.\n\nLOCATION (lock): The table seating area inside the small restaurant hall, lit by daytime light from the windows. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The finger-length, arrowhead-shaped animal-bone necklace remains hanging from Jihu's neck.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 3 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물의 예리한 시선과 뼈 목걸이, 식탁 위 사물들을 자연스럽게 구현하여 지시사항에 가장 충실합니다." }, { "label": "B", "score": 3, "verdict_ko": "지문에 명시되지 않은 배경 인물들을 다수 추가하여 등장인물 제한 규칙을 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "목걸이 줄이 인물의 턱과 목을 뚫고 들어가는 물리적으로 불가능한 형태가 렌더링되어 치명적인 위반입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S21sh11_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S21sh11" } }, "S21sh19": { "input_fingerprint": "e376cbab490793ff", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset transitioning to night.\n\nSHOT TEXT (authoritative, Korean): 맨손으로 붉은 생고기를 꽉 쥔 채 입안 가득 밀어넣고 있는 중간 찰나, 핏물로 붉게 물든 한국인 꼬마 아이들의 입가.\n\nLOCATION (lock): Around the food-laden table in the small restaurant hall. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same table setting, abundant raw meat, warm daytime restaurant light, and food arrangement. Exclude the adults from the immediate frame and show only the children greedily eating the meat by hand.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The children continue eating the raw-meat spread with their bare hands, leaving fresh blood around their mouths and on their hands; Jihu still wears her necklace nearby.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 8, "B": 3, "C": 2 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "어른을 배제하라는 지시를 따랐으며, 참조 이미지의 테이블 세팅을 유지한 채 아이들이 맨손으로 생고기를 입에 밀어넣고 핏물이 묻은 모습을 매우 정확하게 묘사했습니다." }, { "label": "B", "score": 3, "verdict_ko": "어른은 배제했으나, 이전 숏의 붉은 생고기 세팅을 무시하고 테이블 중앙에 불판과 구운 고기를 임의로 생성하여 연속성 및 소품 지시를 크게 위반했습니다." }, { "label": "C", "score": 2, "verdict_ko": "생고기가 아닌 조리된 꼬치구이를 배치했으며, 프레임에서 어른을 제외하라는 명시적 지시를 어기고 우측 및 배경에 어른을 포함하는 치명적 오류를 범했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S21sh16_sel.png" }, { "label": "PROP REFERENCE — 붉은 생고기: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S21sh16" } }, "S22sh2": { "input_fingerprint": "35f9ccbf9a802c93", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset transitioning to night.\n\nSHOT TEXT (authoritative, Korean): 어둠이 내린 섬 마을 길가, 바싹 말라 쪼그라든 꽃잎의 클로즈업.\n\nLOCATION (lock): A roadside flower patch within the island’s coastal village after nightfall. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The flowers across the island have withered as night falls and remain shriveled.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 6, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "해안 마을이라는 배경이 가장 잘 드러나며, 바싹 마른 꽃의 클로즈업 묘사가 지시문에 매우 충실합니다." }, { "label": "A", "score": 6, "verdict_ko": "쪼그라든 꽃잎의 클로즈업은 훌륭하나, 해안가 섬 마을이라는 공간적 맥락이 B에 비해 덜 명확합니다." }, { "label": "C", "score": 4, "verdict_ko": "마른 꽃의 형태는 보이지만 배경이 너무 어두워 마을 길가라는 장소의 분위기를 파악하기 어렵습니다." } ], "refs": [], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "groupbg::해송_숲길": { "input_fingerprint": "352a1905b12e79b2", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "해송_숲길", "tags": [ "S107sh1", "S23sh3", "S29sh11", "S29sh3", "S29sh7", "S30sh1", "S30sh7", "S33sh12", "S33sh13", "S33sh4", "S45sh3", "S45sh6", "S45sh8", "S64sh2", "S67sh1", "S67sh6", "S67sh8", "S69sh3", "S69sh6", "S69sh7", "S74sh2", "S74sh8" ] }, "context_sig": "7ab885356ed33102" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A narrow, unlit trail beneath dense maritime pines, illuminated only by moonlight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 해송 숲길과 숲속: 빽빽한 해송 나무가 가득한 어두운 숲속 오솔길과 비탈. (특징: 빽빽한 해송 나무 기둥과 가지들; 바닥의 수풀과 나뭇잎; 가로등 없는 좁고 거친 흙길)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 숲길 - 밤\n- 실외. 숲길 - 밤\n- 실외. 숲속 - 밤\n- 실외. 숲길 - 낮\n- 실외. 숲길 - 밤\n- 실외. 숲속 - 밤\n- 실외. 숲 속 - 밤\n\nTIME OF DAY (lock): night, moonlit.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A narrow, unlit trail beneath dense maritime pines, illuminated only by moonlight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 해송 숲길과 숲속: 빽빽한 해송 나무가 가득한 어두운 숲속 오솔길과 비탈. (특징: 빽빽한 해송 나무 기둥과 가지들; 바닥의 수풀과 나뭇잎; 가로등 없는 좁고 거친 흙길)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 숲길 - 밤\n- 실외. 숲길 - 밤\n- 실외. 숲속 - 밤\n- 실외. 숲길 - 낮\n- 실외. 숲길 - 밤\n- 실외. 숲속 - 밤\n- 실외. 숲 속 - 밤\n\nTIME OF DAY (lock): night, moonlit.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png", "asset_id": "dded1895-8314-4902-9d6c-855ef980a51c", "input_asset_ids": [ "387a3a7b-93e3-4c2f-a36d-68a1cbd6b195" ], "origin_tag": "S23sh3", "place_text": "A narrow, unlit trail beneath dense maritime pines, illuminated only by moonlight.", "origin_inputs": { "place_text": "A narrow, unlit trail beneath dense maritime pines, illuminated only by moonlight.", "time_of_day_en": "night, moonlit", "conti_asset_id": "387a3a7b-93e3-4c2f-a36d-68a1cbd6b195" } }, "S23sh3::bgfirst_bg": { "input_fingerprint": "a39016a28f51987f", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 달빛을 받아 차갑게 빛나는 고산들의 단호하고 무심한 옆얼굴.\n\nLOCATION (lock): A narrow, unlit trail beneath dense maritime pines, illuminated only by moonlight.\n\nTIME OF DAY (lock): night, moonlit.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera tracks at eye level directly beside 고산들, matching his fast stride and holding a tight perpendicular profile as the forest slips behind him. His coldly lit cheek and set jaw occupy the right two-thirds, leaving a narrow leading-space channel in the direction of travel.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 고산들 in the middle-right of the frame, midground, moves toward forest path ahead; forest path ahead in the middle-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 울창한 해송 나무 숲 (달빛만 비치는 밤) — The trunks pass laterally behind his profile as the camera tracks beside him; used as Moving background layers that convey his uninterrupted pace; 어두운 숲길 (불빛이 없음); used as A narrow channel of leading space beside his face.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 달빛을 받아 차갑게 빛나는 고산들의 단호하고 무심한 옆얼굴.\n\nLOCATION (lock): A narrow, unlit trail beneath dense maritime pines, illuminated only by moonlight.\n\nTIME OF DAY (lock): night, moonlit.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera tracks at eye level directly beside 고산들, matching his fast stride and holding a tight perpendicular profile as the forest slips behind him. His coldly lit cheek and set jaw occupy the right two-thirds, leaving a narrow leading-space channel in the direction of travel.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 고산들 in the middle-right of the frame, midground, moves toward forest path ahead; forest path ahead in the middle-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 울창한 해송 나무 숲 (달빛만 비치는 밤) — The trunks pass laterally behind his profile as the camera tracks beside him; used as Moving background layers that convey his uninterrupted pace; 어두운 숲길 (불빛이 없음); used as A narrow channel of leading space beside his face.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S23sh3__bgfirst_bg.png", "asset_id": "91e60fcb-dc96-469a-839a-5a9dfeebcd34", "input_asset_ids": [ "387a3a7b-93e3-4c2f-a36d-68a1cbd6b195", "dded1895-8314-4902-9d6c-855ef980a51c" ] }, "S23sh3": { "input_fingerprint": "c61191326a912561", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 달빛을 받아 차갑게 빛나는 고산들의 단호하고 무심한 옆얼굴.\n\nLOCATION (lock): A narrow, unlit trail beneath dense maritime pines, illuminated only by moonlight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 달빛을 받아 차갑게 빛나는 고산들의 단호하고 무심한 옆얼굴.\n\nLOCATION (lock): A narrow, unlit trail beneath dense maritime pines, illuminated only by moonlight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 달빛을 받아 차갑게 빛나는 고산들의 단호하고 무심한 옆얼굴.\n\nLOCATION (lock): A narrow, unlit trail beneath dense maritime pines, illuminated only by moonlight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S23sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S23sh3.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "제시된 인물 레퍼런스의 외모와 의상(모자, 가죽 재킷)을 완벽하게 유지하면서, 달빛이 비치는 단호한 옆얼굴이라는 숏 묘사를 매우 사실적으로 잘 구현했습니다." }, { "label": "A", "score": 8, "verdict_ko": "옆얼굴 구도와 밤의 소나무 숲 배경은 훌륭하게 연출되었으나, 인물 레퍼런스의 의상과 모자가 누락되고 정장으로 임의 변경되어 감점되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "제시된 인물 레퍼런스의 외모와 의상(모자, 가죽 재킷)을 완벽하게 유지하면서, 달빛이 비치는 단호한 옆얼굴이라는 숏 묘사를 매우 사실적으로 잘 구현했습니다." }, { "label": "A", "score": 8, "verdict_ko": "옆얼굴 구도와 밤의 소나무 숲 배경은 훌륭하게 연출되었으나, 인물 레퍼런스의 의상과 모자가 누락되고 정장으로 임의 변경되어 감점되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 숲길 로케이션과 달빛 조명을 사실적으로 구현했으며, 레퍼런스 이미지의 의상(가죽 재킷, 모자)을 정확히 유지하면서 텍스트가 요구한 단호하고 무심한 옆얼굴을 완벽하게 포착했습니다." }, { "label": "B", "score": 7, "verdict_ko": "옆얼굴의 구도와 달빛의 묘사는 지시문에 부합하나, 캐릭터 레퍼런스에 주어진 의상을 무시하고 임의로 정장 재킷을 입혀 복장 일관성에서 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "지정된 숲길 로케이션과 달빛 조명을 사실적으로 구현했으며, 레퍼런스 이미지의 의상(가죽 재킷, 모자)을 정확히 유지하면서 텍스트가 요구한 단호하고 무심한 옆얼굴을 완벽하게 포착했습니다." }, { "label": "A", "score": 7, "verdict_ko": "옆얼굴의 구도와 달빛의 묘사는 지시문에 부합하나, 캐릭터 레퍼런스에 주어진 의상을 무시하고 임의로 정장 재킷을 입혀 복장 일관성에서 감점되었습니다." } ] }, "combined": { "totals": { "A": 15, "B": 20 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 15, "B": 20 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "제시된 인물 레퍼런스의 외모와 의상(모자, 가죽 재킷)을 완벽하게 유지하면서, 달빛이 비치는 단호한 옆얼굴이라는 숏 묘사를 매우 사실적으로 잘 구현했습니다." }, { "label": "A", "score": 8, "verdict_ko": "옆얼굴 구도와 밤의 소나무 숲 배경은 훌륭하게 연출되었으나, 인물 레퍼런스의 의상과 모자가 누락되고 정장으로 임의 변경되어 감점되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S23sh3__bgfirst_bg.png", "bg_asset_id": "91e60fcb-dc96-469a-839a-5a9dfeebcd34", "bg_record_key": "S23sh3::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "해송_숲길", "groupbg_asset_id": "dded1895-8314-4902-9d6c-855ef980a51c" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S24sh1::bgfirst_bg": { "input_fingerprint": "9c04817be3b34ee2", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두컴컴한 식당 안, 아주 살짝 열린 방문 틈 사이로 어둠 속을 엿보는 김지후(강수리영)의 불안한 눈동자.\n\nLOCATION (lock): The doorway between a back room and the dark restaurant hall, with weak moonlight entering through the windows.\n\nTIME OF DAY (lock): night, moonlit.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the dining-room side of the barely open doorway, a static eye-level camera peers into the narrow gap and frames 김지후(강수리영)'s anxious eye with only part of her brow and cheek. The two door edges compress her into the center-left sliver, while the dark dining room remains faintly legible beyond the foreground edge.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: left edge of the barely open doorway in the middle-left of the frame, foreground; right edge of the barely open doorway in the middle-right of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 살짝 열린 방문 (아주 살짝 열려 있음) — The camera sees the inner edges of the door and frame, with the opening aimed into the dining room; used as Two close foreground edges form a narrow through-object frame around the visible eye; 식당 내부 (어두컴컴하고 반응하는 사람이 없음); used as Sparse negative space behind the doorway slit.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두컴컴한 식당 안, 아주 살짝 열린 방문 틈 사이로 어둠 속을 엿보는 김지후(강수리영)의 불안한 눈동자.\n\nLOCATION (lock): The doorway between a back room and the dark restaurant hall, with weak moonlight entering through the windows.\n\nTIME OF DAY (lock): night, moonlit.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the dining-room side of the barely open doorway, a static eye-level camera peers into the narrow gap and frames 김지후(강수리영)'s anxious eye with only part of her brow and cheek. The two door edges compress her into the center-left sliver, while the dark dining room remains faintly legible beyond the foreground edge.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: left edge of the barely open doorway in the middle-left of the frame, foreground; right edge of the barely open doorway in the middle-right of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 살짝 열린 방문 (아주 살짝 열려 있음) — The camera sees the inner edges of the door and frame, with the opening aimed into the dining room; used as Two close foreground edges form a narrow through-object frame around the visible eye; 식당 내부 (어두컴컴하고 반응하는 사람이 없음); used as Sparse negative space behind the doorway slit.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S24sh1__bgfirst_bg.png", "asset_id": "2b9243da-4299-46ca-b59f-e5db08ab85e3", "input_asset_ids": [ "5e25d74f-0bfc-4886-8832-a2e56c84c397", "15c006bc-c7ae-4006-8e15-ae5074040f95" ] }, "S24sh1": { "input_fingerprint": "984493bb02387bb8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 어두컴컴한 식당 안, 아주 살짝 열린 방문 틈 사이로 어둠 속을 엿보는 김지후(강수리영)의 불안한 눈동자.\n\nLOCATION (lock): The doorway between a back room and the dark restaurant hall, with weak moonlight entering through the windows. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her arrowhead-shaped bone necklace as she leaves her room that night.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 어두컴컴한 식당 안, 아주 살짝 열린 방문 틈 사이로 어둠 속을 엿보는 김지후(강수리영)의 불안한 눈동자.\n\nLOCATION (lock): The doorway between a back room and the dark restaurant hall, with weak moonlight entering through the windows. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her arrowhead-shaped bone necklace as she leaves her room that night.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 어두컴컴한 식당 안, 아주 살짝 열린 방문 틈 사이로 어둠 속을 엿보는 김지후(강수리영)의 불안한 눈동자.\n\nLOCATION (lock): The doorway between a back room and the dark restaurant hall, with weak moonlight entering through the windows. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her arrowhead-shaped bone necklace as she leaves her room that night.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S24sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S24sh1.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B06.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "살짝 열린 문틈 사이로 엿보는 불안한 눈동자라는 숏의 핵심 프레이밍을 정확하게 묘사했으며, 배경의 주방 구조와 소품도 잘 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지시문이 요구한 문틈 사이의 눈동자 클로즈업 대신 프레임이 불필요하게 넓어졌으며, 카메라를 가리는 앞쪽 구조물의 개연성도 떨어집니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "살짝 열린 문틈 사이로 엿보는 불안한 눈동자라는 숏의 핵심 프레이밍을 정확하게 묘사했으며, 배경의 주방 구조와 소품도 잘 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지시문이 요구한 문틈 사이의 눈동자 클로즈업 대신 프레임이 불필요하게 넓어졌으며, 카메라를 가리는 앞쪽 구조물의 개연성도 떨어집니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "식당 안에서 살짝 열린 방문 틈으로 어둠을 내다보는 카메라 구도와 시선을 정확히 연출했으며, 요구된 목걸이도 잘 묘사되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물이 식당 홀을 등지고 방 안을 들여다보는 구도라 '식당 안 어둠 속을 엿보는' 지시문과 반대되며, 문의 형태도 지시문과 맞지 않습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "식당 안에서 살짝 열린 방문 틈으로 어둠을 내다보는 카메라 구도와 시선을 정확히 연출했으며, 요구된 목걸이도 잘 묘사되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "인물이 식당 홀을 등지고 방 안을 들여다보는 구도라 '식당 안 어둠 속을 엿보는' 지시문과 반대되며, 문의 형태도 지시문과 맞지 않습니다." } ] }, "combined": { "totals": { "A": 9, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 9, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "살짝 열린 문틈 사이로 엿보는 불안한 눈동자라는 숏의 핵심 프레이밍을 정확하게 묘사했으며, 배경의 주방 구조와 소품도 잘 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지시문이 요구한 문틈 사이의 눈동자 클로즈업 대신 프레임이 불필요하게 넓어졌으며, 카메라를 가리는 앞쪽 구조물의 개연성도 떨어집니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B06.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S24sh1__bgfirst_bg.png", "bg_asset_id": "2b9243da-4299-46ca-b59f-e5db08ab85e3", "bg_record_key": "S24sh1::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S24sh3": { "input_fingerprint": "ca072b1c214fe40d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 희미한 달빛만이 비치는 텅 빈 식당 한가운데 홀로 서서 목을 길게 빼고 주변을 살피는 김지후(강수리영)의 옆모습.\n\nLOCATION (lock): The center of the empty restaurant hall at night, barely visible in moonlight filtering through the windows. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains alone in the dark restaurant with the bone necklace still around her neck.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 6, "A": 4 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "레퍼런스의 전통 목재 창호문과 배경 디테일을 정확히 유지하면서, 홀로 서 있는 옆모습의 구도를 훌륭하게 구현했습니다." }, { "label": "C", "score": 6, "verdict_ko": "인물의 포즈와 분위기는 좋으나, 배경에 있는 창호문의 창살 디자인이 레퍼런스와 다르게 묘사되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "레퍼런스에 확립된 전통 창호문 대신 현대식 유리문이 등장하여 필수적인 로케이션 일관성을 크게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S24sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S24sh1" } }, "S25sh2::bgfirst_bg": { "input_fingerprint": "38c7e2d42b22e1d1", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 검은 아가리처럼 뻥 뚫린 동굴 입구 앞에 멈춰 서서 안쪽을 가만히 들여다보는 고산들의 뒷모습.\n\nLOCATION (lock): The forest clearing directly before a large natural cave mouth reinforced with wooden posts.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Directly behind 고산들 at shoulder height, the static camera looks past his motionless back and between his shoulders into the open cave. His figure occupies the lower central third while the cave mouth and its added wooden supports remain fully legible around him, sharing his hesitant line of sight.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 자연동굴 입구 (열려 있고 안쪽이 끝없이 깜깜함) — The entrance faces the camera behind 고산들, exposing the unlit cave interior; used as The central background void directly aligned with 고산들's gaze; 입구의 나무 기둥 (사람이 덧댄 것처럼 설치되어 있음) — Their inward-facing sides border the cave mouth on either side of 고산들; used as An enclosing structural frame around the cave opening.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 검은 아가리처럼 뻥 뚫린 동굴 입구 앞에 멈춰 서서 안쪽을 가만히 들여다보는 고산들의 뒷모습.\n\nLOCATION (lock): The forest clearing directly before a large natural cave mouth reinforced with wooden posts.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Directly behind 고산들 at shoulder height, the static camera looks past his motionless back and between his shoulders into the open cave. His figure occupies the lower central third while the cave mouth and its added wooden supports remain fully legible around him, sharing his hesitant line of sight.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 자연동굴 입구 (열려 있고 안쪽이 끝없이 깜깜함) — The entrance faces the camera behind 고산들, exposing the unlit cave interior; used as The central background void directly aligned with 고산들's gaze; 입구의 나무 기둥 (사람이 덧댄 것처럼 설치되어 있음) — Their inward-facing sides border the cave mouth on either side of 고산들; used as An enclosing structural frame around the cave opening.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S25sh2__bgfirst_bg.png", "asset_id": "863a8220-40a9-4ef5-a40d-ce0e9632da38", "input_asset_ids": [ "e6bc229d-60d4-45e9-a662-301f38503ec9", "dad15ae4-e0d6-4247-8fc4-392b8ffcb643", "cf15bf22-d776-4059-aabd-5729e3547256" ] }, "S25sh2": { "input_fingerprint": "1f43e7cd6e35775e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 검은 아가리처럼 뻥 뚫린 동굴 입구 앞에 멈춰 서서 안쪽을 가만히 들여다보는 고산들의 뒷모습.\n\nLOCATION (lock): The forest clearing directly before a large natural cave mouth reinforced with wooden posts. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 검은 아가리처럼 뻥 뚫린 동굴 입구 앞에 멈춰 서서 안쪽을 가만히 들여다보는 고산들의 뒷모습.\n\nLOCATION (lock): The forest clearing directly before a large natural cave mouth reinforced with wooden posts. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 검은 아가리처럼 뻥 뚫린 동굴 입구 앞에 멈춰 서서 안쪽을 가만히 들여다보는 고산들의 뒷모습.\n\nLOCATION (lock): The forest clearing directly before a large natural cave mouth reinforced with wooden posts. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S25sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S25sh2.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L29B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_natural_cave_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물의 뒷모습과 의상을 사실적으로 구현했으며, 구조물 레퍼런스의 디테일(상자와 밧줄)을 배경에 더 잘 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "지시문의 상황과 장소는 무난하게 연출했으나, 구조물 레퍼런스의 세부 요소 반영이 A에 비해 부족합니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물의 뒷모습과 의상을 사실적으로 구현했으며, 구조물 레퍼런스의 디테일(상자와 밧줄)을 배경에 더 잘 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "지시문의 상황과 장소는 무난하게 연출했으나, 구조물 레퍼런스의 세부 요소 반영이 A에 비해 부족합니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "인물의 뒷모습과 의상 디테일을 잘 구현했으며, 구조물 레퍼런스에 있는 소품(밧줄, 상자, 장작)을 프레임에 반영하려는 시도가 돋보입니다." }, { "label": "A", "score": 5, "verdict_ko": "인물의 뒷모습은 지시대로 묘사되었으나, 구조물 레퍼런스의 형태나 디테일을 전혀 반영하지 않고 위치 레퍼런스만 그대로 차용했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물의 뒷모습과 의상 디테일을 잘 구현했으며, 구조물 레퍼런스에 있는 소품(밧줄, 상자, 장작)을 프레임에 반영하려는 시도가 돋보입니다." }, { "label": "B", "score": 5, "verdict_ko": "인물의 뒷모습은 지시대로 묘사되었으나, 구조물 레퍼런스의 형태나 디테일을 전혀 반영하지 않고 위치 레퍼런스만 그대로 차용했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 11 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 11 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물의 뒷모습과 의상을 사실적으로 구현했으며, 구조물 레퍼런스의 디테일(상자와 밧줄)을 배경에 더 잘 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "지시문의 상황과 장소는 무난하게 연출했으나, 구조물 레퍼런스의 세부 요소 반영이 A에 비해 부족합니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L29B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_natural_cave_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S25sh2__bgfirst_bg.png", "bg_asset_id": "863a8220-40a9-4ef5-a40d-ce0e9632da38", "bg_record_key": "S25sh2::bgfirst_bg", "chain_winner": true, "authority": "plate", "seed_attached": true }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S25sh4": { "input_fingerprint": "3f429f99b6b8e054", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 칠흑 같은 동굴 내부를 노려보며 미간을 찌푸린 채 복잡한 심경을 드러내는 고산들의 무거운 얼굴.\n\nLOCATION (lock): At the threshold of the forest cave, facing into its completely dark interior. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same timber-braced cave mouth, surrounding forest, moonlit darkness, and deep black interior. Exclude the rear-view composition and frame the man's conflicted face at the entrance.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 4 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물의 얼굴을 가장 가깝게 프레이밍하여 '무거운 얼굴'에 집중하라는 숏 지시를 가장 잘 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "얼굴을 강조하라는 지시와 달리 프레임이 넓어져 불필요하게 신체를 더 많이 보여줍니다." }, { "label": "C", "score": 4, "verdict_ko": "세 후보 중 숏 크기가 가장 넓어 인물의 표정에 집중하라는 프레이밍 우선순위를 크게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S25sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S25sh2" } }, "S26sh3::bgfirst_bg": { "input_fingerprint": "7b60a927620459af", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 창고의 거친 나무 문 표면에 한쪽 귀를 바짝 갖다 댄 채 숨을 죽인 김지후(강수리영)의 상체.\n\nLOCATION (lock): The small yard immediately outside a rough wooden shed behind the restaurant at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-torso height beside 김지후(강수리영), the static camera sits perpendicular to the storage-room door and tightens around her shoulder, cheek, and ear pressed to its rough surface. She occupies the left two-thirds, while the door and a narrow trace of escaping light remain on the right to establish what she is listening through.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 창고의 거친 나무 문 (문 밖으로 불빛이 새어나옴) — Its outer face lies nearly parallel to the image plane, with the narrow opening positioned beyond her ear; used as The contact surface for her ear and the dominant vertical plane beside her profile; 식당 뒤뜰 (밤); used as A restrained background sliver establishing the exterior location.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 창고의 거친 나무 문 표면에 한쪽 귀를 바짝 갖다 댄 채 숨을 죽인 김지후(강수리영)의 상체.\n\nLOCATION (lock): The small yard immediately outside a rough wooden shed behind the restaurant at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-torso height beside 김지후(강수리영), the static camera sits perpendicular to the storage-room door and tightens around her shoulder, cheek, and ear pressed to its rough surface. She occupies the left two-thirds, while the door and a narrow trace of escaping light remain on the right to establish what she is listening through.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 창고의 거친 나무 문 (문 밖으로 불빛이 새어나옴) — Its outer face lies nearly parallel to the image plane, with the narrow opening positioned beyond her ear; used as The contact surface for her ear and the dominant vertical plane beside her profile; 식당 뒤뜰 (밤); used as A restrained background sliver establishing the exterior location.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S26sh3__bgfirst_bg.png", "asset_id": "bf13f875-5548-4c0c-bf32-42de03188de0", "input_asset_ids": [ "61cf28ae-b401-4a1a-afd9-3b2bc59ef086", "47476310-e478-4051-9653-24a34609c0eb", "69cb1065-4089-4f87-94f0-4059c3fec052" ] }, "S26sh3": { "input_fingerprint": "1ffe30cc92eb3446", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 창고의 거친 나무 문 표면에 한쪽 귀를 바짝 갖다 댄 채 숨을 죽인 김지후(강수리영)의 상체.\n\nLOCATION (lock): The small yard immediately outside a rough wooden shed behind the restaurant at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her arrowhead-shaped bone necklace while listening at the shed door.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 창고의 거친 나무 문 표면에 한쪽 귀를 바짝 갖다 댄 채 숨을 죽인 김지후(강수리영)의 상체.\n\nLOCATION (lock): The small yard immediately outside a rough wooden shed behind the restaurant at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her arrowhead-shaped bone necklace while listening at the shed door.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 창고의 거친 나무 문 표면에 한쪽 귀를 바짝 갖다 댄 채 숨을 죽인 김지후(강수리영)의 상체.\n\nLOCATION (lock): The small yard immediately outside a rough wooden shed behind the restaurant at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her arrowhead-shaped bone necklace while listening at the shed door.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S26sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S26sh3.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L31B03.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_restaurant_compound_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "귀를 문에 대지 않고 살짝 열린 문틈으로 안을 들여다보고 있어 텍스트의 핵심 행동을 따르지 않았습니다." }, { "label": "B", "score": 8, "verdict_ko": "거친 나무 문에 한쪽 귀를 바짝 갖다 댄 상체의 모습을 텍스트의 지시대로 정확하게 연출했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "귀를 문에 대지 않고 살짝 열린 문틈으로 안을 들여다보고 있어 텍스트의 핵심 행동을 따르지 않았습니다." }, { "label": "B", "score": 8, "verdict_ko": "거친 나무 문에 한쪽 귀를 바짝 갖다 댄 상체의 모습을 텍스트의 지시대로 정확하게 연출했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "B는 장소와 구조물 레퍼런스(철문과 금속 손잡이)를 완벽히 재현했으나, 재킷이 누락되었고 귀를 문에 밀착하는 동작 묘사가 다소 부족합니다." }, { "label": "A", "score": 5, "verdict_ko": "A는 의상을 잘 재현했으나, 텍스트와 레퍼런스 충돌 시 레퍼런스를 우선하라는 지시를 무시하고 철문을 나무 문으로 변경하여 우선순위 3을 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "B는 장소와 구조물 레퍼런스(철문과 금속 손잡이)를 완벽히 재현했으나, 재킷이 누락되었고 귀를 문에 밀착하는 동작 묘사가 다소 부족합니다." }, { "label": "B", "score": 5, "verdict_ko": "A는 의상을 잘 재현했으나, 텍스트와 레퍼런스 충돌 시 레퍼런스를 우선하라는 지시를 무시하고 철문을 나무 문으로 변경하여 우선순위 3을 위반했습니다." } ] }, "combined": { "totals": { "A": 11, "B": 13 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 13 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "귀를 문에 대지 않고 살짝 열린 문틈으로 안을 들여다보고 있어 텍스트의 핵심 행동을 따르지 않았습니다." }, { "label": "B", "score": 8, "verdict_ko": "거친 나무 문에 한쪽 귀를 바짝 갖다 댄 상체의 모습을 텍스트의 지시대로 정확하게 연출했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L31B03.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_restaurant_compound_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S26sh3__bgfirst_bg.png", "bg_asset_id": "bf13f875-5548-4c0c-bf32-42de03188de0", "bg_record_key": "S26sh3::bgfirst_bg", "chain_winner": false, "authority": "plate", "seed_attached": true }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S26sh5": { "input_fingerprint": "7b375374f9bce4da", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 문틈 너머, 붉은 촛불빛에 기괴한 제단의 윤곽이 드러나 있는 창고 내부의 음산한 전경.\n\nLOCATION (lock): A narrow view through the rear-yard shed door into a candlelit ritual space. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inside the shed, the candlelit altar remains set with its ritual objects; Jihu still wears her bone necklace outside the door.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 4, "A": 4 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "구조물 외형(골함석 소재, 창문 없음)은 '구조물 외형' 사진을, 주변 환경은 '위치' 사진을 따르라는 까다로운 우선순위 규칙을 가장 정확하게 준수했습니다." }, { "label": "C", "score": 4, "verdict_ko": "제단의 기괴한 분위기는 잘 묘사되었으나, 우선순위 규칙을 무시하고 위치 사진 속 구조물(시멘트 블록 및 창문)을 그대로 차용하여 지정 조건을 위반했습니다." }, { "label": "A", "score": 4, "verdict_ko": "우선순위 지침을 어기고 위치 사진의 구조물(시멘트 벽돌, 방범창) 형태를 그대로 렌더링하는 치명적인 오류를 범했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L31B03.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_restaurant_compound_sel.png" } ], "critique_skipped": true, "ref_mode": "플레이트+seed만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only" }, "S27sh3": { "input_fingerprint": "41c4e85287724fae", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 십수 개의 촛불 사이로 연기가 피어오르는 제단 위, 검붉은 피가 가득 담긴 사발과 검은 염소 머리의 기괴한 모습.\n\nLOCATION (lock): Inside the rear-yard ritual shed, around a candlelit altar filled with incense smoke. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The severed black-goat head and the bowl filled with dark-red blood remain arranged on the candlelit altar. Jihu still wears her bone necklace.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 6, "A": 4 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "촛불, 피가 담긴 사발, 잘린 염소 머리 등 묘사된 제단의 모습을 요구된 샷 크기에 맞춰 가장 사실적인 구도로 구현했습니다." }, { "label": "C", "score": 6, "verdict_ko": "제시된 제단의 주요 요소들을 잘 포함했으나, 염소 머리가 살아있는 것처럼 반듯하게 놓여 있어 잘린 머리의 묘사가 다소 인위적입니다." }, { "label": "A", "score": 4, "verdict_ko": "제단의 클로즈업 대신 너무 넓은 화각을 선택했으며, 창고 내부와 외부 마당이 비정상적으로 섞인 구조 왜곡이 발생했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S26sh5_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S26sh5" }, "lane_policy": "share_plan_prev_bgonly" }, "S27sh5": { "input_fingerprint": "cff0d8bfd5a713a8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 넋을 잃은 김지후(강수리영)의 등 뒤, 어두운 천장 구석에 기괴한 자세로 매달려 있는 인간형 실루엣의 일부.\n\nLOCATION (lock): A shadowed upper corner where the wall meets the ceiling inside the candlelit ritual shed. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same candlelit altar, goat head, blood bowl, rising incense, and sinister shed materials. Exclude any prior empty altar-only framing and add the woman below with the hidden figure clinging to the ceiling behind her.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old wooden spear with the antler-like tip remains displayed above the goat head, while the goat head and blood-filled bowl stay on the altar. Jihu still wears the matching bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "주문한 앵글과 인물의 넋을 잃은 표정을 잘 구현했으며, 장소의 목재 재질과 제단 디테일을 충실히 유지했습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 텍스트에 명시된 필수 전경 인물(김지후)이 프레임에서 완전히 누락되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "천장이 현대식 흡음 타일로 렌더링되어 장소 고정을 위반했고, 뼈 목걸이가 후드 뒷면에 가로로 붙어 있는 물리적 오류가 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S27sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S27sh3" } }, "S28sh3": { "input_fingerprint": "69114639357252b2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 새카만 산길 입구를 올려다보며 어깨를 한껏 움츠린 채 굳어버린 김지후(강수리영)의 뒷모습.\n\nLOCATION (lock): The edge of the restaurant’s rear yard where the village path rises toward the dark forest. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rough shed exterior, vegetable garden, yard surfaces, and moonless darkness. Exclude the ear-to-door action and show the woman hesitating toward the black mountain path.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her arrowhead-shaped bone necklace as she leaves the shed and considers entering the forest.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "움츠린 뒷모습을 훌륭하게 연출했으며, 헛간과 채소밭 등 레퍼런스의 배경 디테일을 가장 정확하게 유지했습니다." }, { "label": "B", "score": 5, "verdict_ko": "뒷모습과 포즈는 적절히 구현되었으나, 이전 샷에 있던 채소밭과 주변 소품들이 누락되어 장소 일관성이 떨어집니다." }, { "label": "A", "score": 3, "verdict_ko": "뒷모습임에도 불구하고 목걸이가 등 뒤에 매달려 있는 물리적으로 불가능한 위치 오류가 발생했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S26sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S26sh3" } }, "S29sh3::bgfirst_bg": { "input_fingerprint": "2a26fd6d3d08b733", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 숲길, 표면이 불룩하게 튀어나온 커다란 포댓자루 두 개가 실린 리어카를 끄는 방향으로 몸을 기울여 앞으로 한 발을 내디딘 중간 동작, 한국인 마을 청년들의 짙은 실루엣.\n\nLOCATION (lock): A narrow nighttime trail deep within the maritime-pine forest.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level beside the concealed position, the camera tracks perpendicular to the Korean village youths as they pass across the frame with the cart and its two bulging sacks. Their dark bodies are caught mid-stride in the middle ground, leaning into the pull, while the cart remains fully legible behind them and the forest path supplies scale.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 마을 청년#1 in the middle-right of the frame, midground, moves toward far side of the forest path; cart with two sacks in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: cart (Moving along the path) — Its side faces the camera while its front points along the youths' direction of travel; used as Travels with the youths as the principal load-bearing shape in the tracking composition; two large sacks (Loaded on the cart and moving from within) — Their bulging sides face the camera from atop the cart; used as Creates two distinct irregular masses on the cart without revealing their contents; forest path (Dark) — The path crosses laterally through the frame along the group's travel axis; used as Provides the transverse corridor through which the group passes.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 숲길, 표면이 불룩하게 튀어나온 커다란 포댓자루 두 개가 실린 리어카를 끄는 방향으로 몸을 기울여 앞으로 한 발을 내디딘 중간 동작, 한국인 마을 청년들의 짙은 실루엣.\n\nLOCATION (lock): A narrow nighttime trail deep within the maritime-pine forest.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level beside the concealed position, the camera tracks perpendicular to the Korean village youths as they pass across the frame with the cart and its two bulging sacks. Their dark bodies are caught mid-stride in the middle ground, leaning into the pull, while the cart remains fully legible behind them and the forest path supplies scale.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 마을 청년#1 in the middle-right of the frame, midground, moves toward far side of the forest path; cart with two sacks in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: cart (Moving along the path) — Its side faces the camera while its front points along the youths' direction of travel; used as Travels with the youths as the principal load-bearing shape in the tracking composition; two large sacks (Loaded on the cart and moving from within) — Their bulging sides face the camera from atop the cart; used as Creates two distinct irregular masses on the cart without revealing their contents; forest path (Dark) — The path crosses laterally through the frame along the group's travel axis; used as Provides the transverse corridor through which the group passes.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S29sh3__bgfirst_bg.png", "asset_id": "7ba01a99-79f2-4b3a-b9e7-cb6534bff090", "input_asset_ids": [ "4aa11ff2-bcf7-492d-b11e-a03bf7799a40", "dded1895-8314-4902-9d6c-855ef980a51c" ] }, "S29sh3": { "input_fingerprint": "d9d4bc2b2df6bc34", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 숲길, 표면이 불룩하게 튀어나온 커다란 포댓자루 두 개가 실린 리어카를 끄는 방향으로 몸을 기울여 앞으로 한 발을 내디딘 중간 동작, 한국인 마을 청년들의 짙은 실루엣.\n\nLOCATION (lock): A narrow nighttime trail deep within the maritime-pine forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The two large, moving sacks remain loaded on the youths' handcart, and Jihu still wears her bone necklace while watching them.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 숲길, 표면이 불룩하게 튀어나온 커다란 포댓자루 두 개가 실린 리어카를 끄는 방향으로 몸을 기울여 앞으로 한 발을 내디딘 중간 동작, 한국인 마을 청년들의 짙은 실루엣.\n\nLOCATION (lock): A narrow nighttime trail deep within the maritime-pine forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The two large, moving sacks remain loaded on the youths' handcart, and Jihu still wears her bone necklace while watching them.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 숲길, 표면이 불룩하게 튀어나온 커다란 포댓자루 두 개가 실린 리어카를 끄는 방향으로 몸을 기울여 앞으로 한 발을 내디딘 중간 동작, 한국인 마을 청년들의 짙은 실루엣.\n\nLOCATION (lock): A narrow nighttime trail deep within the maritime-pine forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The two large, moving sacks remain loaded on the youths' handcart, and Jihu still wears her bone necklace while watching them.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S29sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S29sh3.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 포댓자루: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" }, { "label": "PROP REFERENCE — 리어카: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 포댓자루: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" }, { "label": "PROP REFERENCE — 리어카: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 없는 인물을 임의로 추가하여 심각한 규정을 위반했으며, 지정된 리어카 소품의 형태를 전혀 반영하지 못했습니다." }, { "label": "B", "score": 7, "verdict_ko": "주요 인물이 완전한 실루엣으로 표현되지 않은 점은 다소 아쉬우나, 지정된 리어카와 포댓자루 소품, 캐릭터의 복장 및 동작을 충실하게 재현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 없는 인물을 임의로 추가하여 심각한 규정을 위반했으며, 지정된 리어카 소품의 형태를 전혀 반영하지 못했습니다." }, { "label": "B", "score": 7, "verdict_ko": "주요 인물이 완전한 실루엣으로 표현되지 않은 점은 다소 아쉬우나, 지정된 리어카와 포댓자루 소품, 캐릭터의 복장 및 동작을 충실하게 재현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "소품 레퍼런스(리어카, 포댓자루)를 충실히 반영했으나, 주 인물이 짙은 실루엣으로 처리되지 않고 미는 동작처럼 연출된 점이 아쉽습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 텍스트에 없는 인물(지후)을 임의로 등장시켜 치명적 위반이며, 리어카의 디자인도 레퍼런스와 전혀 다릅니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "소품 레퍼런스(리어카, 포댓자루)를 충실히 반영했으나, 주 인물이 짙은 실루엣으로 처리되지 않고 미는 동작처럼 연출된 점이 아쉽습니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 없는 인물(지후)을 임의로 등장시켜 치명적 위반이며, 리어카의 디자인도 레퍼런스와 전혀 다릅니다." } ] }, "combined": { "totals": { "A": 6, "B": 13 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 13 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 없는 인물을 임의로 추가하여 심각한 규정을 위반했으며, 지정된 리어카 소품의 형태를 전혀 반영하지 못했습니다." }, { "label": "B", "score": 7, "verdict_ko": "주요 인물이 완전한 실루엣으로 표현되지 않은 점은 다소 아쉬우나, 지정된 리어카와 포댓자루 소품, 캐릭터의 복장 및 동작을 충실하게 재현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 포댓자루: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" }, { "label": "PROP REFERENCE — 리어카: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S29sh3__bgfirst_bg.png", "bg_asset_id": "7ba01a99-79f2-4b3a-b9e7-cb6534bff090", "bg_record_key": "S29sh3::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "해송_숲길", "groupbg_asset_id": "dded1895-8314-4902-9d6c-855ef980a51c" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S29sh7": { "input_fingerprint": "c21be80c72bbd17c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 귓가에 입술을 댄 채 반대편 손가락을 자신의 입술에 바짝 붙인 김여상의 굳은 표정.\n\nLOCATION (lock): Dense undergrowth beside the forest trail, concealed from the passing villagers. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark pine path, dense undergrowth, cart tracks, and sparse moonlight. Exclude the cart pullers from the close frame and show the fisherman silencing the woman.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu continues to wear her bone necklace while Yeosang restrains her from making a sound.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 김여상 (40대 중반 한국인 남성, 평범한 중년 얼굴, 단정한 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 4 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "귓가에 입술을 대고 반대편 손가락을 자신의 입술에 바짝 붙인 행동 지문을 완벽하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "인물의 구도는 좋으나 남성의 손가락이 입술이 아닌 뺨을 향해 있어 세부 행동 묘사가 부정확합니다." }, { "label": "A", "score": 4, "verdict_ko": "귀에 입술을 대지 않았으며, 남성 캐릭터가 지정된 외양이 아닌 이전 컷의 수레꾼 모자를 쓰고 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S29sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김여상: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S29sh3" } }, "S29sh11": { "input_fingerprint": "661d29837c64ffe6", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 숲속으로 멀어져 가는 김여상의 뒷모습을 원망과 결의가 섞인 눈빛으로 노려보는 김지후(강수리영)의 얼굴.\n\nLOCATION (lock): The dark forest trail where it disappears deeper among the pines. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dense pine forest, narrow path, and low moonlight. Exclude the youths and cart from the immediate frame, retaining only the fisherman receding into the woods.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu still wears her bone necklace as she decides to follow Yeosang.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지후가 멀어져 가는 남자를 노려보는 시선 방향과 상황을 정확히 연출했으며, 남자의 모자(이전 컷 타 인물의 소품) 오류는 하위 우선순위라 가장 우수함." }, { "label": "B", "score": 5, "verdict_ko": "지후가 남자를 등지고 완전히 다른 곳을 응시하고 있어 '남자의 뒷모습을 노려본다'는 핵심 행동 지시(우선순위 3)를 크게 위반함." }, { "label": "A", "score": 3, "verdict_ko": "이전 컷에 존재하지 않던 커다란 바구니를 남자의 등에 임의로 추가하여 '창조된 사물'을 금지하는 하드 위반에 해당함." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S29sh7_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S29sh7" } }, "S30sh1": { "input_fingerprint": "4c3e55098840114e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 숲길 한가운데 멈춰 선 리어카에서 검붉은 피가 배어 나오는 커다란 포댓자루를 어깨에 번쩍 짊어진 한국인 마을 청년들의 땀에 젖은 상체.\n\nLOCATION (lock): A trackless section of deep forest where the cart can go no farther. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same pine trunks, narrow trail, undergrowth, and dark moonlit atmosphere. Exclude the earlier distant cart arrangement and show the youths lifting the blood-soaked sacks.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The youths have taken the two sacks off the handcart and now carry them on their shoulders; dark-red liquid continues leaking from them. Jihu still wears her bone necklace while following.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "'땀에 젖은 상체' 지문을 맨몸으로 정확히 구현했으며, 레퍼런스와 일치하는 리어카와 포댓자루의 형태를 훌륭하게 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "전경의 인물이 상의를 그대로 입고 있어 '땀에 젖은 상체'라는 묘사와 어긋나며, 배경의 리어카 형태도 레퍼런스와 다릅니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 텍스트에 없는 인물(여성)이 우측에 등장하여 등장인물 제한을 지시한 프롬프트 규칙을 심각하게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S29sh11_sel.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 포댓자루: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" }, { "label": "PROP REFERENCE — 리어카: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S29sh11" } }, "S30sh7": { "input_fingerprint": "f068c0804e853cec", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수풀에 엎드린 채 숨을 꾹 참고 서로의 눈을 맞추며 얼어붙은 김지후(강수리영)와 김여상의 옆얼굴.\n\nLOCATION (lock): Low vegetation off the deep-forest route, providing cover from the villagers ahead. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark trail, dense brush, pine trunks, and blood-marked route. Exclude the standing youths and cart from the foreground, showing the two pursuers flattened behind the shrubs.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu and Yeosang remain hidden flat behind the brush while the youths carry the two blood-leaking sacks onward; Jihu still wears her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 김여상 (40대 중반 한국인 남성, 평범한 중년 얼굴, 단정한 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "샷 텍스트의 측면 구도와 덤불 속 은폐 설정을 잘 구현했으며, 지정된 인물과 의상(빨간 재킷)을 가장 정확히 반영했습니다." }, { "label": "B", "score": 5, "verdict_ko": "구도와 엎드린 자세는 적절하나, 김여상의 의상이 레퍼런스(빨간 재킷)와 완전히 다른 어두운 색상으로 렌더링되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "샷 텍스트에 없는 배경 인물들을 추가하여 등장인물 제한 규칙을 위반했으며, 인물들이 덤불이 아닌 개방된 길 위에 노출되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S30sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김여상: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S30sh1" } }, "groupbg::숲속_염소_축사": { "input_fingerprint": "e9eb94402b00a2a9", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "숲속_염소_축사", "tags": [ "S109sh1", "S31sh2", "S31sh3", "S58sh2" ] }, "context_sig": "e866422cbe0ad3e5" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: The narrow exterior passage running through a forest goat enclosure at night.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n숲속 염소 축사: 숲 깊은 곳에 위치한 울타리 형태의 가건물. 다수의 덩치 큰 뿔 달린 검은 염소들이 몰려 있다. (특징: 나무 울타리와 기둥; 건초와 배설물이 섞인 바닥; 검은 털에 두 뿔이 달린 염소 떼)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 숲속 염소 축사 - 밤\n- 검은 염소 수십 마리가 축사 안에 몰려있다.\n- 실외. 염소 축사 - 낮\n- 실외. 염소 축사 - 밤\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: The narrow exterior passage running through a forest goat enclosure at night.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n숲속 염소 축사: 숲 깊은 곳에 위치한 울타리 형태의 가건물. 다수의 덩치 큰 뿔 달린 검은 염소들이 몰려 있다. (특징: 나무 울타리와 기둥; 건초와 배설물이 섞인 바닥; 검은 털에 두 뿔이 달린 염소 떼)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 숲속 염소 축사 - 밤\n- 검은 염소 수십 마리가 축사 안에 몰려있다.\n- 실외. 염소 축사 - 낮\n- 실외. 염소 축사 - 밤\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_숲속_염소_축사_f6a67e.png", "asset_id": "b1ef5a8e-92a6-4bc1-89ca-de320c8d4a75", "input_asset_ids": [ "5553f879-eeb6-4a01-b18f-bdd9046f3be9" ], "origin_tag": "S31sh2", "place_text": "The narrow exterior passage running through a forest goat enclosure at night.", "origin_inputs": { "place_text": "The narrow exterior passage running through a forest goat enclosure at night.", "time_of_day_en": "night", "conti_asset_id": "5553f879-eeb6-4a01-b18f-bdd9046f3be9" } }, "S31sh2::bgfirst_bg": { "input_fingerprint": "2efc4bf3ccb65721", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 축사 사이의 좁은 길에서 한쪽 다리를 들어 올려 앞으로 뻗은 mid-stride 자세로, 핏물이 떨어지는 포댓자루를 어깨에 멘 한국인 마을 청년들의 뒷모습.\n\nLOCATION (lock): The narrow exterior passage running through a forest goat enclosure at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height several steps behind the Korean village youths, the tracking frame widens enough to catch 마을 청년#1 with one leg extended forward while the group enters the narrow route between the enclosures. His back and the sack on his shoulder stay centered, with falling dark-red liquid legible below the load and black goats held to both sides as spatial context.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 마을 청년#1 in the middle-center of the frame, midground, moves toward narrow route between the enclosures; narrow route between goat enclosures in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: goat enclosures (Holding dozens of black goats) — Their inward-facing sides flank the path on the left and right; used as Forms the confined travel corridor around the moving group; black goats (Gathered inside the enclosures); used as Crowds both sides of the youths' route and deepens the sense of confinement; large sack (Moving violently from within and dripping dark-red liquid) — Its rear and lower side face the following camera; used as Rides over the carrier's shoulder as the secondary focal mass.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 축사 사이의 좁은 길에서 한쪽 다리를 들어 올려 앞으로 뻗은 mid-stride 자세로, 핏물이 떨어지는 포댓자루를 어깨에 멘 한국인 마을 청년들의 뒷모습.\n\nLOCATION (lock): The narrow exterior passage running through a forest goat enclosure at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height several steps behind the Korean village youths, the tracking frame widens enough to catch 마을 청년#1 with one leg extended forward while the group enters the narrow route between the enclosures. His back and the sack on his shoulder stay centered, with falling dark-red liquid legible below the load and black goats held to both sides as spatial context.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 마을 청년#1 in the middle-center of the frame, midground, moves toward narrow route between the enclosures; narrow route between goat enclosures in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: goat enclosures (Holding dozens of black goats) — Their inward-facing sides flank the path on the left and right; used as Forms the confined travel corridor around the moving group; black goats (Gathered inside the enclosures); used as Crowds both sides of the youths' route and deepens the sense of confinement; large sack (Moving violently from within and dripping dark-red liquid) — Its rear and lower side face the following camera; used as Rides over the carrier's shoulder as the secondary focal mass.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S31sh2__bgfirst_bg.png", "asset_id": "d39d11ef-0485-4a0a-a0ee-445683f0b57f", "input_asset_ids": [ "5553f879-eeb6-4a01-b18f-bdd9046f3be9", "b1ef5a8e-92a6-4bc1-89ca-de320c8d4a75" ] }, "S31sh2": { "input_fingerprint": "087821eeb4f0396a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 축사 사이의 좁은 길에서 한쪽 다리를 들어 올려 앞으로 뻗은 mid-stride 자세로, 핏물이 떨어지는 포댓자루를 어깨에 멘 한국인 마을 청년들의 뒷모습.\n\nLOCATION (lock): The narrow exterior passage running through a forest goat enclosure at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The youths continue carrying the two blood-leaking sacks through the goat enclosure, with the contents still struggling. Jihu still wears her bone necklace while following.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 축사 사이의 좁은 길에서 한쪽 다리를 들어 올려 앞으로 뻗은 mid-stride 자세로, 핏물이 떨어지는 포댓자루를 어깨에 멘 한국인 마을 청년들의 뒷모습.\n\nLOCATION (lock): The narrow exterior passage running through a forest goat enclosure at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The youths continue carrying the two blood-leaking sacks through the goat enclosure, with the contents still struggling. Jihu still wears her bone necklace while following.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 축사 사이의 좁은 길에서 한쪽 다리를 들어 올려 앞으로 뻗은 mid-stride 자세로, 핏물이 떨어지는 포댓자루를 어깨에 멘 한국인 마을 청년들의 뒷모습.\n\nLOCATION (lock): The narrow exterior passage running through a forest goat enclosure at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The youths continue carrying the two blood-leaking sacks through the goat enclosure, with the contents still struggling. Jihu still wears her bone necklace while following.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S31sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S31sh2.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 포댓자루: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_숲속_염소_축사_f6a67e.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 포댓자루: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 인물(여성 등)을 추가했으며, 지정된 마을 청년 레퍼런스의 복장과 외형을 전혀 따르지 않았습니다." }, { "label": "B", "score": 7, "verdict_ko": "지정된 캐릭터 레퍼런스의 외형과 복장을 정확히 반영했으며, 텍스트에 묘사된 독특한 다리 자세와 피가 떨어지는 자루를 충실히 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 인물(여성 등)을 추가했으며, 지정된 마을 청년 레퍼런스의 복장과 외형을 전혀 따르지 않았습니다." }, { "label": "B", "score": 7, "verdict_ko": "지정된 캐릭터 레퍼런스의 외형과 복장을 정확히 반영했으며, 텍스트에 묘사된 독특한 다리 자세와 피가 떨어지는 자루를 충실히 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물 레퍼런스의 복장(비니, 후드집업 등)을 완벽히 일치시켰으며, 다리를 들어 앞으로 뻗은 구체적인 자세와 피 흘리는 포댓자루를 잘 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "다리를 뒤로 들어올려 '앞으로 뻗은'이라는 동작 지침을 어겼으며, 레퍼런스와 일치하지 않는 복장 및 제약에 위배되는 인물(지후)을 포함시켰습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "인물 레퍼런스의 복장(비니, 후드집업 등)을 완벽히 일치시켰으며, 다리를 들어 앞으로 뻗은 구체적인 자세와 피 흘리는 포댓자루를 잘 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "다리를 뒤로 들어올려 '앞으로 뻗은'이라는 동작 지침을 어겼으며, 레퍼런스와 일치하지 않는 복장 및 제약에 위배되는 인물(지후)을 포함시켰습니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 인물(여성 등)을 추가했으며, 지정된 마을 청년 레퍼런스의 복장과 외형을 전혀 따르지 않았습니다." }, { "label": "B", "score": 7, "verdict_ko": "지정된 캐릭터 레퍼런스의 외형과 복장을 정확히 반영했으며, 텍스트에 묘사된 독특한 다리 자세와 피가 떨어지는 자루를 충실히 구현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_숲속_염소_축사_f6a67e.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 포댓자루: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S31sh2__bgfirst_bg.png", "bg_asset_id": "d39d11ef-0485-4a0a-a0ee-445683f0b57f", "bg_record_key": "S31sh2::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "숲속_염소_축사", "groupbg_asset_id": "b1ef5a8e-92a6-4bc1-89ca-de320c8d4a75" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S31sh3": { "input_fingerprint": "37bc2d2eed791dc8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 청년의 어깨 위에서 포댓자루 안의 무언가가 강하게 밖으로 밀어내어, 자루 표면이 기괴한 형체로 팽팽하게 왜곡된 고정된 순간의 클로즈업.\n\nLOCATION (lock): Within the outdoor goat-pen passage, focused on a sack carried above the penned animals. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same goat-pen passage, dark fencing, moonlight, and the continuing blood-stained sack. Exclude the carrier's full body and isolate the sack surface being distorted from within.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The living contents continue distorting the sack from within, and dark-red liquid remains soaked through it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 5, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 클로즈업 샷에 가장 부합하며 자루 표면이 기괴하게 돌출되는 모습을 잘 포착했습니다." }, { "label": "C", "score": 5, "verdict_ko": "요구된 클로즈업보다 프레임이 넓게 잡혀 인물의 상반신이 너무 많이 노출되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "어깨 위가 아닌 등 한가운데에 자루가 물리적으로 불가능하게 떠 있어 심각한 구조적 오류가 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S31sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 포댓자루: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S31sh2" } }, "S32sh12": { "input_fingerprint": "6238738b0582d70f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 바닥의 포댓자루 밖으로 상체가 반쯤 튀어나온 채 기괴하게 뒤틀려 멈춘 흐릿한 인간 형체의 실루엣\n\nLOCATION (lock): The dim central work floor of a makeshift forest slaughterhouse, beside opened sacks and butchering equipment. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The victim's upper body protrudes halfway from a sack on the slaughterhouse floor and remains twisted motionless, while the lower body stays enclosed inside the sack; Jihu still wears her bone necklace outside the window.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 5 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물을 실제 인간으로 묘사하고, 포댓자루의 상태와 창밖의 지후(뼈 목걸이 착용) 등 프롬프트의 세부 조건을 가장 충실히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "인물을 이목구비가 없는 추상적 덩어리로 묘사하여 '모든 캐릭터는 온전한 인간이어야 한다'는 핵심 규칙을 위반했습니다." }, { "label": "C", "score": 5, "verdict_ko": "포댓자루에 들어간 인물은 표현되었으나, 창밖의 지후와 뼈 목걸이가 누락되었고 기괴하게 뒤틀린 실루엣의 느낌이 부족합니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B01.png" } ], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S32sh13": { "input_fingerprint": "4139ee113ad10586", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 일말의 망설임 없이 무시무시한 도축용 칼을 허공으로 높이 치켜든 마을 청년#1의 굳은 상체\n\nLOCATION (lock): The killing area inside the dark makeshift slaughterhouse, under minimal nighttime light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rough slaughterhouse floor, hanging tools, blood containers, and dim interior light. Exclude the half-emerged victim from prominence and show the youth raising the butcher knife.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The captive remains exposed on the slaughterhouse floor beneath the raised butcher's blade, and Jihu still wears her bone necklace while witnessing it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "이전 샷의 연속성(창밖의 지후 위치, 청년의 비니 복장)을 정확히 유지하며, 지시된 청년의 상체와 높이 든 칼을 충실히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "청년의 얼굴이 잘려 상체 프레이밍 지시를 벗어났으며, 창밖에 있어야 할 지후가 실내로 이동하고 희생자가 과도하게 부각되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "청년의 상체 프레이밍은 맞췄으나 레퍼런스의 비니가 누락되었고, 지후가 실내로 들어와 이전 샷과의 위치 연속성이 깨졌습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S32sh12_sel.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 식칼·도축용 칼·의식용 칼: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S32sh12" } }, "S32sh17": { "input_fingerprint": "f07896a2b642f41c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 피투성이 얼굴로 창문을 노려보는 짐승 같은 형상의 마을 청년#1의 기괴한 표정\n\nLOCATION (lock): Inside the slaughterhouse facing its broken, dusty observation window. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same crude slaughterhouse, hanging blades, bloody floor, and darkness broken by the window. Exclude the victim and knife-swing instant, focusing on the bloodied youth staring toward the window.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The struck captive now lies motionless where it was butchered, with fresh blood splashed around it. Jihu still wears her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "A": 5, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시문대로 희생자를 프레임에서 제외하고 피투성이가 된 청년의 기괴하고 짐승 같은 표정을 가장 정확하게 연출했습니다." }, { "label": "A", "score": 5, "verdict_ko": "희생자를 제외하라는 지시를 어기고 이전 샷에 없던 거대한 동물 사체를 전경에 임의로 추가하여 감점되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "'피투성이 얼굴'과 '기괴한 표정'이라는 핵심 지시를 전혀 반영하지 않고 평범한 상태로 묘사했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S32sh13_sel.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S32sh13" } }, "S33sh4::bgfirst_bg": { "input_fingerprint": "099130f914930717", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 김여상이 사라진 언덕 아래를 내려다보며 경악한 채 우뚝 선 김지후(강수리영)의 굳은 전신\n\nLOCATION (lock): The edge of a steep forest slope beside the nighttime escape route.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At full-body distance beside and slightly behind 김지후(강수리영), hold completely static near the top of the slope and aim downward past her. Her halted body occupies the center-left while the descending ground opens beneath her, leaving the place where 김여상 vanished conspicuously empty.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: slope below where 김여상 disappeared in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 언덕 아래 경사면 (김여상이 굴러 떨어져 시야에서 사라진 상태) — The slope falls away from the camera through the lower and central frame; used as The downward visual anchor beyond 수리영 and the empty destination of her shocked gaze; 잔가지와 수풀 (Crowding the sides of the forest route); used as Frames 수리영's silhouette and preserves the forest pursuit setting.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 김여상이 사라진 언덕 아래를 내려다보며 경악한 채 우뚝 선 김지후(강수리영)의 굳은 전신\n\nLOCATION (lock): The edge of a steep forest slope beside the nighttime escape route.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At full-body distance beside and slightly behind 김지후(강수리영), hold completely static near the top of the slope and aim downward past her. Her halted body occupies the center-left while the descending ground opens beneath her, leaving the place where 김여상 vanished conspicuously empty.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: slope below where 김여상 disappeared in the lower-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 언덕 아래 경사면 (김여상이 굴러 떨어져 시야에서 사라진 상태) — The slope falls away from the camera through the lower and central frame; used as The downward visual anchor beyond 수리영 and the empty destination of her shocked gaze; 잔가지와 수풀 (Crowding the sides of the forest route); used as Frames 수리영's silhouette and preserves the forest pursuit setting.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S33sh4__bgfirst_bg.png", "asset_id": "cde860a4-b3ee-400d-a2f9-831daf7dc0bb", "input_asset_ids": [ "e51c4800-b2f4-48a0-91d4-c87b1ac16b71", "dded1895-8314-4902-9d6c-855ef980a51c" ] }, "S33sh4": { "input_fingerprint": "49d9fd6b92d084e4", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 김여상이 사라진 언덕 아래를 내려다보며 경악한 채 우뚝 선 김지후(강수리영)의 굳은 전신\n\nLOCATION (lock): The edge of a steep forest slope beside the nighttime escape route. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang has fallen out of sight down the slope and does not return to Jihu; Jihu still wears her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 김여상이 사라진 언덕 아래를 내려다보며 경악한 채 우뚝 선 김지후(강수리영)의 굳은 전신\n\nLOCATION (lock): The edge of a steep forest slope beside the nighttime escape route. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang has fallen out of sight down the slope and does not return to Jihu; Jihu still wears her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 김여상이 사라진 언덕 아래를 내려다보며 경악한 채 우뚝 선 김지후(강수리영)의 굳은 전신\n\nLOCATION (lock): The edge of a steep forest slope beside the nighttime escape route. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang has fallen out of sight down the slope and does not return to Jihu; Jihu still wears her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S33sh4__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S33sh4.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '전신' 샷의 스케일과 장소를 정확히 구현했으나, 뒷모습 구도로 인해 요구된 경악한 표정이 직접적으로 드러나지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "표정은 잘 드러나나 '전신' 샷 요구를 어기고 프레임을 좁혔으며, 지시문에 없는 손전등을 임의로 추가하여 우선순위 규정을 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '전신' 샷의 스케일과 장소를 정확히 구현했으나, 뒷모습 구도로 인해 요구된 경악한 표정이 직접적으로 드러나지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "표정은 잘 드러나나 '전신' 샷 요구를 어기고 프레임을 좁혔으며, 지시문에 없는 손전등을 임의로 추가하여 우선순위 규정을 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지문에서 요구한 '전신' 프레이밍을 정확하게 구현했으며 언덕 아래를 내려다보는 자세를 잘 표현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "인물의 하반신이 잘려 샷 텍스트에 명시된 '전신' 구도라는 최우선 프레이밍 조건을 충족하지 못했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문에서 요구한 '전신' 프레이밍을 정확하게 구현했으며 언덕 아래를 내려다보는 자세를 잘 표현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 하반신이 잘려 샷 텍스트에 명시된 '전신' 구도라는 최우선 프레이밍 조건을 충족하지 못했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 7 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 7 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '전신' 샷의 스케일과 장소를 정확히 구현했으나, 뒷모습 구도로 인해 요구된 경악한 표정이 직접적으로 드러나지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "표정은 잘 드러나나 '전신' 샷 요구를 어기고 프레임을 좁혔으며, 지시문에 없는 손전등을 임의로 추가하여 우선순위 규정을 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S33sh4__bgfirst_bg.png", "bg_asset_id": "cde860a4-b3ee-400d-a2f9-831daf7dc0bb", "bg_record_key": "S33sh4::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "해송_숲길", "groupbg_asset_id": "dded1895-8314-4902-9d6c-855ef980a51c" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S33sh12": { "input_fingerprint": "33ee020eb8fe474c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 고산들의 손에 목이 졸린 채 두 눈이 커진 김지후(강수리영)의 고통스러운 얼굴\n\nLOCATION (lock): An open footpath where two forest routes meet beneath the pines at night. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark pine trunks, narrow path, and tangled brush. Exclude the cart and sacks, and show the man gripping the woman's throat at the open trail.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul keeps Jihu suspended by the throat as she struggles, with her bone necklace still around her neck.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 고산들 (각성 상태) (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 3, "B": 7, "C": 5 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "두 인물이 물리적 접촉 없이 나란히 서 있어 목을 조르는 핵심 행동과 연출이 전혀 반영되지 않았습니다." }, { "label": "B", "score": 7, "verdict_ko": "어깨너머 숏으로 지후의 커진 눈과 산들의 손을 텍스트에 맞게 강조했으나, 맨목 대신 옷깃을 잡고 비니가 누락되었습니다." }, { "label": "C", "score": 5, "verdict_ko": "지후의 표정은 나타나지만 산들의 얼굴이 프레임에 크게 잡혀 초점이 분산되며, 역시 목이 아닌 옷깃을 쥐고 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S33sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 고산들 (각성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S33sh4" } }, "S33sh13": { "input_fingerprint": "753974888bc5368e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 눈동자가 황금빛으로 번뜩이고 동공이 직사각형으로 변한 고산들의 무시무시한 얼굴 클로즈업\n\nLOCATION (lock): The same moonlit forest footpath at the edge of dense undergrowth. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cold moonlit forest palette and dark pine backdrop. Exclude the fleeing figures and focus only on the man's transformed golden eyes and rectangular pupils.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul's eyes remain golden with rectangular pupils while his hand continues crushing Jihu's throat; her bone necklace remains on her neck until she loses consciousness.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (각성 상태) (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "A": 5, "C": 5 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 프레이밍(얼굴 클로즈업)을 완벽히 따르며, 황금빛 눈동자와 뚜렷한 직사각형 동공을 가장 정확하게 묘사했습니다." }, { "label": "A", "score": 5, "verdict_ko": "얼굴 클로즈업과 황금빛 눈동자는 잘 표현되었으나, 동공이 직사각형이라기보다는 파충류나 고양이의 세로 선창 모양에 가깝습니다." }, { "label": "C", "score": 5, "verdict_ko": "클로즈업과 배경은 일치하지만, 눈동자의 동공이 요구된 직사각형이 아닌 일반적인 짐승의 세로 동공으로 묘사되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S33sh12_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들 (각성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S33sh12" } }, "groupbg::달리는_승용차_객실": { "input_fingerprint": "6a4a704084842990", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "달리는_승용차_객실", "tags": [ "S34sh2", "S34sh4" ] }, "context_sig": "11c871c84bf58093" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: The driver’s seat inside a moving compact sedan on a rural highway in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n구형 아반테 내부: 현대 소형 세단의 운전석. 스마트폰 거치대와 앞유리창. (특징: 내비게이션용 거치된 스마트폰; 운전대; 햇빛이 들어오는 앞유리)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실내. 달리는 아반테 - 낮\n- 부아아앙- 지방 국도를 질주하는 아반테.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: The driver’s seat inside a moving compact sedan on a rural highway in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n구형 아반테 내부: 현대 소형 세단의 운전석. 스마트폰 거치대와 앞유리창. (특징: 내비게이션용 거치된 스마트폰; 운전대; 햇빛이 들어오는 앞유리)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실내. 달리는 아반테 - 낮\n- 부아아앙- 지방 국도를 질주하는 아반테.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_달리는_승용차_객실_5ca19f.png", "asset_id": "9d17f929-ee14-4268-b99b-e5dad371fdb1", "input_asset_ids": [ "8a2e4c46-105d-40a7-aa34-bda695608e7e" ], "origin_tag": "S34sh2", "place_text": "The driver’s seat inside a moving compact sedan on a rural highway in daylight.", "origin_inputs": { "place_text": "The driver’s seat inside a moving compact sedan on a rural highway in daylight.", "time_of_day_en": "day", "conti_asset_id": "8a2e4c46-105d-40a7-aa34-bda695608e7e" } }, "S34sh2::bgfirst_bg": { "input_fingerprint": "5914bdc48ba587df", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한쪽 팔에 깁스를 한 채 인상을 잔뜩 찌푸리고 운전대를 꽉 쥔 정인우의 상체\n\nLOCATION (lock): The driver’s seat inside a moving compact sedan on a rural highway in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the moving car from the front passenger side, finish the dolly-in at 정인우's chest height and an arm's-length distance. Frame his tense upper body in three-quarter profile with his cast arm and gripping hand readable around the steering wheel, while his worried face remains oriented toward the road rather than the camera.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 운전대 (Held tightly while the car is moving) — The rim is seen obliquely from the passenger side with In-woo's hands gripping it; used as Foreground action anchor linking In-woo's rigid hands to the moving vehicle; 앞유리 (The car is traveling along a rural road) — The camera sees the rural road ahead through it from the passenger-side cabin angle; used as Provides the forward driving context behind In-woo's profile; 핸드폰 (Being used for a hands-free call to 지후) — Its screen side is angled away from full view in this shot, with the call interface not yet clearly readable; used as The next focal destination as the camera prepares to tilt and shift away from In-woo.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한쪽 팔에 깁스를 한 채 인상을 잔뜩 찌푸리고 운전대를 꽉 쥔 정인우의 상체\n\nLOCATION (lock): The driver’s seat inside a moving compact sedan on a rural highway in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the moving car from the front passenger side, finish the dolly-in at 정인우's chest height and an arm's-length distance. Frame his tense upper body in three-quarter profile with his cast arm and gripping hand readable around the steering wheel, while his worried face remains oriented toward the road rather than the camera.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 운전대 (Held tightly while the car is moving) — The rim is seen obliquely from the passenger side with In-woo's hands gripping it; used as Foreground action anchor linking In-woo's rigid hands to the moving vehicle; 앞유리 (The car is traveling along a rural road) — The camera sees the rural road ahead through it from the passenger-side cabin angle; used as Provides the forward driving context behind In-woo's profile; 핸드폰 (Being used for a hands-free call to 지후) — Its screen side is angled away from full view in this shot, with the call interface not yet clearly readable; used as The next focal destination as the camera prepares to tilt and shift away from In-woo.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S34sh2__bgfirst_bg.png", "asset_id": "11adc4ad-aac8-48fc-95aa-40cad968e847", "input_asset_ids": [ "8a2e4c46-105d-40a7-aa34-bda695608e7e", "9d17f929-ee14-4268-b99b-e5dad371fdb1" ] }, "S34sh2": { "input_fingerprint": "7c240b2de04b6117", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한쪽 팔에 깁스를 한 채 인상을 잔뜩 찌푸리고 운전대를 꽉 쥔 정인우의 상체\n\nLOCATION (lock): The driver’s seat inside a moving compact sedan on a rural highway in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo's previously injured arm is now immobilized in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한쪽 팔에 깁스를 한 채 인상을 잔뜩 찌푸리고 운전대를 꽉 쥔 정인우의 상체\n\nLOCATION (lock): The driver’s seat inside a moving compact sedan on a rural highway in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo's previously injured arm is now immobilized in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한쪽 팔에 깁스를 한 채 인상을 잔뜩 찌푸리고 운전대를 꽉 쥔 정인우의 상체\n\nLOCATION (lock): The driver’s seat inside a moving compact sedan on a rural highway in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo's previously injured arm is now immobilized in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S34sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S34sh2.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_달리는_승용차_객실_5ca19f.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 로케이션 레퍼런스의 차량 내부(대시보드, 폰 거치대 등)를 완벽하게 재현했으며, 깁스를 한 상태로 운전대를 꽉 쥐고 찌푸린 표정의 지시를 충실히 따랐습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 표정과 깁스 설정은 지시를 잘 따랐으나, 차량 내부 구조와 대시보드가 지정된 로케이션 레퍼런스와 전혀 일치하지 않아 감점되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 로케이션 레퍼런스의 차량 내부(대시보드, 폰 거치대 등)를 완벽하게 재현했으며, 깁스를 한 상태로 운전대를 꽉 쥐고 찌푸린 표정의 지시를 충실히 따랐습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 표정과 깁스 설정은 지시를 잘 따랐으나, 차량 내부 구조와 대시보드가 지정된 로케이션 레퍼런스와 전혀 일치하지 않아 감점되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "차량 내부 대시보드와 창밖 배경 등 로케이션 레퍼런스를 완벽하게 일치시켰으며, 찌푸린 표정과 깁스 설정도 잘 구현되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "인물의 표정과 깁스를 한 채 운전대를 쥔 묘사는 좋으나, 차량 내부 형태와 배경이 로케이션 레퍼런스와 전혀 다릅니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "차량 내부 대시보드와 창밖 배경 등 로케이션 레퍼런스를 완벽하게 일치시켰으며, 찌푸린 표정과 깁스 설정도 잘 구현되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 표정과 깁스를 한 채 운전대를 쥔 묘사는 좋으나, 차량 내부 형태와 배경이 로케이션 레퍼런스와 전혀 다릅니다." } ] }, "combined": { "totals": { "A": 14, "B": 8 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 8 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 로케이션 레퍼런스의 차량 내부(대시보드, 폰 거치대 등)를 완벽하게 재현했으며, 깁스를 한 상태로 운전대를 꽉 쥐고 찌푸린 표정의 지시를 충실히 따랐습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 표정과 깁스 설정은 지시를 잘 따랐으나, 차량 내부 구조와 대시보드가 지정된 로케이션 레퍼런스와 전혀 일치하지 않아 감점되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_달리는_승용차_객실_5ca19f.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S34sh2__bgfirst_bg.png", "bg_asset_id": "11adc4ad-aac8-48fc-95aa-40cad968e847", "bg_record_key": "S34sh2::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "달리는_승용차_객실", "groupbg_asset_id": "9d17f929-ee14-4268-b99b-e5dad371fdb1" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S34sh4": { "input_fingerprint": "1d217a1a80a35b18", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 핸드폰 화면 통화 기록에 '김지후'라는 이름만 빼곡하게 떠 있는 클로즈업\n\nLOCATION (lock): The front cabin of the moving sedan, focused on a phone near the driver. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues driving with one arm in a cast.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 6, "A": 4 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "신체 일부를 배제하라는 지시를 완벽히 준수하였고, '김지후'로 가득 찬 통화 기록 클로즈업을 사실적으로 연출했습니다." }, { "label": "C", "score": 6, "verdict_ko": "인물 배제 및 클로즈업 지시를 잘 따랐으나, 화면 내 일부 텍스트에 무의미한 문자들이 포함되어 아쉽습니다." }, { "label": "A", "score": 4, "verdict_ko": "통화 기록 화면은 잘 구현했으나, 깁스한 팔이 프레임에 노출되어 인물 배제 지시를 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S34sh2_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S34sh2" }, "lane_policy": "share_plan_prev_bgonly" }, "groupbg::꿈속_오솔길": { "input_fingerprint": "210f1266a9ad4279", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "꿈속_오솔길", "tags": [ "S35sh1", "S35sh6", "S35sh8" ] }, "context_sig": "db7a789eece02aaa" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A sunlit, unnaturally still footpath within the island forest, presented as a dream.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n꿈속 금월도 숲속 오솔길: 거리가 비논리적으로 고정되는 기이한 느낌의 숲속 길. (특징: 끝없이 이어지는 듯한 고정된 원근감의 오솔길; 비정상적으로 억눌린 움직임의 공간감)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 숲속 오솔길 - 낮(수리영의 꿈)\n- 수리영이 오솔길에 홀로 서 있다.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A sunlit, unnaturally still footpath within the island forest, presented as a dream.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n꿈속 금월도 숲속 오솔길: 거리가 비논리적으로 고정되는 기이한 느낌의 숲속 길. (특징: 끝없이 이어지는 듯한 고정된 원근감의 오솔길; 비정상적으로 억눌린 움직임의 공간감)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 숲속 오솔길 - 낮(수리영의 꿈)\n- 수리영이 오솔길에 홀로 서 있다.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_꿈속_오솔길_2dac26.png", "asset_id": "9ac4c0c6-1703-4499-9fc1-95b0a78df8b9", "input_asset_ids": [ "3d8a08b3-a368-4f7f-b67a-283a64f7cf0f" ], "origin_tag": "S35sh1", "place_text": "A sunlit, unnaturally still footpath within the island forest, presented as a dream.", "origin_inputs": { "place_text": "A sunlit, unnaturally still footpath within the island forest, presented as a dream.", "time_of_day_en": "day", "conti_asset_id": "3d8a08b3-a368-4f7f-b67a-283a64f7cf0f" } }, "S35sh1::bgfirst_bg": { "input_fingerprint": "9fa389699705ba1d", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 햇살이 비치는 고요한 오솔길 한가운데 홀로 선 김지후(강수리영)의 멍한 전신\n\nLOCATION (lock): A sunlit, unnaturally still footpath within the island forest, presented as a dream.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level on the lower section of the path, hold a static wide view several body lengths from 김지후(강수리영), facing her and the path extending beyond. She is isolated at center with her full body rigid and unfocused, while the open path above her creates a deep, unreachable axis toward the unseen figure ahead.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: path extending uphill beyond 수리영 in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 숲속 오솔길 (Quiet and sunlit in the dream) — The path runs from the lower camera position through 수리영 and continues upward behind her; used as Central depth axis that establishes both 수리영's isolation and the impossible distance she will attempt to cross; 주변 숲 (Bordering both sides of the path); used as Side enclosure that leaves the path as the only visual route forward.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 햇살이 비치는 고요한 오솔길 한가운데 홀로 선 김지후(강수리영)의 멍한 전신\n\nLOCATION (lock): A sunlit, unnaturally still footpath within the island forest, presented as a dream.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level on the lower section of the path, hold a static wide view several body lengths from 김지후(강수리영), facing her and the path extending beyond. She is isolated at center with her full body rigid and unfocused, while the open path above her creates a deep, unreachable axis toward the unseen figure ahead.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: path extending uphill beyond 수리영 in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 숲속 오솔길 (Quiet and sunlit in the dream) — The path runs from the lower camera position through 수리영 and continues upward behind her; used as Central depth axis that establishes both 수리영's isolation and the impossible distance she will attempt to cross; 주변 숲 (Bordering both sides of the path); used as Side enclosure that leaves the path as the only visual route forward.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S35sh1__bgfirst_bg.png", "asset_id": "3bb0a092-9109-4808-a065-cd1679882434", "input_asset_ids": [ "3d8a08b3-a368-4f7f-b67a-283a64f7cf0f", "9ac4c0c6-1703-4499-9fc1-95b0a78df8b9" ] }, "S35sh1": { "input_fingerprint": "61081fb9fbb3306c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 햇살이 비치는 고요한 오솔길 한가운데 홀로 선 김지후(강수리영)의 멍한 전신\n\nLOCATION (lock): A sunlit, unnaturally still footpath within the island forest, presented as a dream. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Within the dream, Jihu appears with the same bone necklace she wore before losing consciousness.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 햇살이 비치는 고요한 오솔길 한가운데 홀로 선 김지후(강수리영)의 멍한 전신\n\nLOCATION (lock): A sunlit, unnaturally still footpath within the island forest, presented as a dream. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Within the dream, Jihu appears with the same bone necklace she wore before losing consciousness.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 햇살이 비치는 고요한 오솔길 한가운데 홀로 선 김지후(강수리영)의 멍한 전신\n\nLOCATION (lock): A sunlit, unnaturally still footpath within the island forest, presented as a dream. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Within the dream, Jihu appears with the same bone necklace she wore before losing consciousness.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S35sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S35sh1.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_꿈속_오솔길_2dac26.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 장소에서 뼈 목걸이를 착용한 인물의 전신이 프레임 안에 온전히 담겨 샷 묘사를 정확히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "'전신' 샷을 요구한 프롬프트와 달리 인물의 무릎 아래가 프레임에서 잘려나가 지시를 온전히 따르지 못했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 장소에서 뼈 목걸이를 착용한 인물의 전신이 프레임 안에 온전히 담겨 샷 묘사를 정확히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "'전신' 샷을 요구한 프롬프트와 달리 인물의 무릎 아래가 프레임에서 잘려나가 지시를 온전히 따르지 못했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "요구된 '전신' 앵글을 정확히 구현하여 머리부터 발끝까지 온전히 담아냈으며, 의상과 목걸이, 배경의 분위기까지 지시문과 훌륭하게 일치합니다." }, { "label": "A", "score": 4, "verdict_ko": "'전신'을 촬영하라는 명확한 지시에도 불구하고 무릎 아래가 프레임에서 잘려나가 샷 크기(staging) 기준을 충족하지 못했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "요구된 '전신' 앵글을 정확히 구현하여 머리부터 발끝까지 온전히 담아냈으며, 의상과 목걸이, 배경의 분위기까지 지시문과 훌륭하게 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "'전신'을 촬영하라는 명확한 지시에도 불구하고 무릎 아래가 프레임에서 잘려나가 샷 크기(staging) 기준을 충족하지 못했습니다." } ] }, "combined": { "totals": { "A": 16, "B": 9 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 16, "B": 9 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 장소에서 뼈 목걸이를 착용한 인물의 전신이 프레임 안에 온전히 담겨 샷 묘사를 정확히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "'전신' 샷을 요구한 프롬프트와 달리 인물의 무릎 아래가 프레임에서 잘려나가 지시를 온전히 따르지 못했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_꿈속_오솔길_2dac26.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S35sh1__bgfirst_bg.png", "bg_asset_id": "3bb0a092-9109-4808-a065-cd1679882434", "bg_record_key": "S35sh1::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "꿈속_오솔길", "groupbg_asset_id": "9ac4c0c6-1703-4499-9fc1-95b0a78df8b9" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S35sh6": { "input_fingerprint": "646cf12fa2e7db2d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 두 눈동자가 섬뜩한 황금색으로 빛나며 직사각형 동공으로 변한 강민숙(홍은미)의 기괴한 얼굴 클로즈업\n\nLOCATION (lock): The upper stretch of the dreamlike forest path, separated from the daughter by an impossible distance. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same unnaturally calm sunlit trail, green forest, and dreamlike distance. Exclude the younger woman from the close frame and show the mother's transformed eyes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): In Jihu's dream, Minsook's eyes have turned golden with rectangular pupils and her body is transforming into black smoke.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (꿈속 각성 상태) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 8, "A": 6, "B": 5 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "캐릭터 레퍼런스의 헤어스타일과 이목구비를 완벽히 유지했으며, 지시된 직사각형 동공과 검은 연기 효과를 클로즈업 샷에 가장 충실하게 구현함." }, { "label": "A", "score": 6, "verdict_ko": "지시된 특수 효과(황금색 눈, 검은 연기)와 클로즈업 구도는 좋으나, 레퍼런스와 달리 머리를 묶어 캐릭터 일관성이 떨어짐." }, { "label": "B", "score": 5, "verdict_ko": "헤어스타일이 레퍼런스와 일치하지 않으며, 동공이 직사각형보다는 고양이 눈(세로 슬릿)에 가깝게 표현됨." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S35sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미) (꿈속 각성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S35sh1" } }, "S35sh8": { "input_fingerprint": "a38244b7d0b58a4b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 허공을 향해 비명을 지르듯 입을 크게 벌린 김지후(강수리영)의 경악한 얼굴\n\nLOCATION (lock): The lower portion of the sunlit dream path where the trapped daughter stands alone. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same sunlit dream trail, still forest, and uncanny spatial separation. Exclude the mother's figure from the immediate frame and isolate the younger woman's silent scream.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains immobilized within the dream as Minsook's smoky form disappears.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "'허공을 향해'라는 텍스트에 맞게 시선을 위로 두고 있으며, 요구된 '얼굴' 중심의 클로즈업 프레이밍을 가장 완벽하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "얼굴 위주의 프레이밍은 훌륭하지만 시선이 정면을 향하고 있어 '허공을 향해'라는 지시를 온전히 따르지 못했습니다." }, { "label": "A", "score": 3, "verdict_ko": "인물의 외형은 가장 정확하지만, '얼굴'에 집중하라는 텍스트를 무시하고 전신 풀샷을 렌더링하여 프레이밍 우선순위를 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S35sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S35sh6" } }, "S36sh5::bgfirst_bg": { "input_fingerprint": "9adeaae9074297d1", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 허전한 목 부위를 향해 황급히 손을 가져다 댄 김지후(강수리영)의 당황한 손\n\nLOCATION (lock): A neatly arranged back room of the restaurant, warmed by daylight entering through its window.\n\nTIME OF DAY (lock): day, warm sunlight.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Place the camera close beside 김지후(강수리영)'s upper torso at neck level, tracking only enough to keep her hurried hand and the bare neck area centered in a tight close frame. Her fingers enter from the lower edge and press against the empty space where the necklace should be, while her face remains partially cropped at the upper edge to make the absence immediately legible.\n- FRAMING SCALE: close-up\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 허전한 목 부위를 향해 황급히 손을 가져다 댄 김지후(강수리영)의 당황한 손\n\nLOCATION (lock): A neatly arranged back room of the restaurant, warmed by daylight entering through its window.\n\nTIME OF DAY (lock): day, warm sunlight.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Place the camera close beside 김지후(강수리영)'s upper torso at neck level, tracking only enough to keep her hurried hand and the bare neck area centered in a tight close frame. Her fingers enter from the lower edge and press against the empty space where the necklace should be, while her face remains partially cropped at the upper edge to make the absence immediately legible.\n- FRAMING SCALE: close-up\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S36sh5__bgfirst_bg.png", "asset_id": "47199cf4-3247-42ab-94cb-93669145374d", "input_asset_ids": [ "0f068fb8-f760-49a3-a6da-c7b5ff8950a5", "4d4446ff-89c9-42cc-96cf-fd8633cbfe19" ] }, "S36sh5": { "input_fingerprint": "7b0f17c925eca0a7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, warm sunlight.\n\nSHOT TEXT (authoritative, Korean): 허전한 목 부위를 향해 황급히 손을 가져다 댄 김지후(강수리영)의 당황한 손\n\nLOCATION (lock): A neatly arranged back room of the restaurant, warmed by daylight entering through its window. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu's arrowhead-shaped bone necklace is now missing from her neck; she remains lying in the neatly arranged back room after waking.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, warm sunlight.\n\nSHOT TEXT (authoritative, Korean): 허전한 목 부위를 향해 황급히 손을 가져다 댄 김지후(강수리영)의 당황한 손\n\nLOCATION (lock): A neatly arranged back room of the restaurant, warmed by daylight entering through its window. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu's arrowhead-shaped bone necklace is now missing from her neck; she remains lying in the neatly arranged back room after waking.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, warm sunlight.\n\nSHOT TEXT (authoritative, Korean): 허전한 목 부위를 향해 황급히 손을 가져다 댄 김지후(강수리영)의 당황한 손\n\nLOCATION (lock): A neatly arranged back room of the restaurant, warmed by daylight entering through its window. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu's arrowhead-shaped bone necklace is now missing from her neck; she remains lying in the neatly arranged back room after waking.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S36sh5__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S36sh5.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L21B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'누워있는' 상태와 목을 감싸는 '당황한 손'에 집중한 클로즈업 구도, 식당 내부가 보이는 배경까지 지시문을 매우 충실히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "인물의 얼굴과 배경의 분위기는 좋으나, 완전히 앉아있는 자세로 묘사되어 '누워있는' 상태 지시를 어겼으며 구도가 손보다 얼굴에 더 집중되어 있습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'누워있는' 상태와 목을 감싸는 '당황한 손'에 집중한 클로즈업 구도, 식당 내부가 보이는 배경까지 지시문을 매우 충실히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "인물의 얼굴과 배경의 분위기는 좋으나, 완전히 앉아있는 자세로 묘사되어 '누워있는' 상태 지시를 어겼으며 구도가 손보다 얼굴에 더 집중되어 있습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "목걸이가 사라진 목을 짚는 손과 당황한 표정을 잘 포착했으며, 지정된 공간의 채광과 분위기를 자연스럽게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "목을 짚는 손가락의 해부학적 구조가 다소 어색하고, 인물의 얼굴이 잘려 당황한 감정이 온전히 전달되지 않습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "목걸이가 사라진 목을 짚는 손과 당황한 표정을 잘 포착했으며, 지정된 공간의 채광과 분위기를 자연스럽게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "목을 짚는 손가락의 해부학적 구조가 다소 어색하고, 인물의 얼굴이 잘려 당황한 감정이 온전히 전달되지 않습니다." } ] }, "combined": { "totals": { "A": 11, "B": 12 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 12 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'누워있는' 상태와 목을 감싸는 '당황한 손'에 집중한 클로즈업 구도, 식당 내부가 보이는 배경까지 지시문을 매우 충실히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "인물의 얼굴과 배경의 분위기는 좋으나, 완전히 앉아있는 자세로 묘사되어 '누워있는' 상태 지시를 어겼으며 구도가 손보다 얼굴에 더 집중되어 있습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L21B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S36sh5__bgfirst_bg.png", "bg_asset_id": "47199cf4-3247-42ab-94cb-93669145374d", "bg_record_key": "S36sh5::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S36sh7": { "input_fingerprint": "5a9f467a10a568cb", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, warm sunlight.\n\nSHOT TEXT (authoritative, Korean): 열린 방문 틈으로 고개를 내민 채 환하게 미소 짓는 강영란(흰 머리 여자)의 넉넉한 얼굴\n\nLOCATION (lock): The open doorway of the restaurant’s tidy back room in warm morning light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without the bone necklace as Youngran enters the room.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지문에 지정된 인물만 등장하여 환하게 웃는 표정을 잘 살렸으며, 방 밖에서 안으로 고개를 내미는 올바른 공간 연출을 보여줍니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 인물을 단독으로 묘사했으나, 방 안(매트리스 쪽)에서 바깥으로 고개를 내미는 구도여서 방으로 들어온다는 설정과 물리적으로 모순됩니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 다른 인물(지후)의 뒷모습을 화면에 포함시켜 인물 추가 금지 규칙을 명백히 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S36sh5_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S36sh5" } }, "S37sh3": { "input_fingerprint": "dee60d26eb0bcc4a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 열린 문 너머로 평범한 식당 물품과 낚시 도구들만 가득 쌓여 있는 창고 내부의 전경.\n\nLOCATION (lock): Inside the rear-yard storage shed, now an ordinary daylight storeroom packed with restaurant and fishing supplies. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The shed has been cleared of the previous night's altar, goat head, blood bowl, candles, and spear, leaving only ordinary restaurant and fishing supplies. Jihu remains without her necklace.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 5 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "열린 문을 통해 창고 내부를 보여달라는 카메라 앵글과 프레이밍 지시를 가장 잘 따랐으며, 식당 및 낚시 물품 묘사가 우수합니다." }, { "label": "B", "score": 3, "verdict_ko": "창고 내부가 아닌 외부를 바라보는 구도로 촬영되어 '창고 내부의 전경'이라는 핵심 지시를 위반했습니다." }, { "label": "C", "score": 5, "verdict_ko": "내부를 보여주지만, 레퍼런스 이미지의 카메라 프레이밍(외부 LPG 가스통과 창문 구도)을 지나치게 모방하여 아쉽습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L31B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_restaurant_compound_sel.png" } ], "critique_skipped": true, "ref_mode": "플레이트+seed만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only" }, "S37sh6::bgfirst_bg": { "input_fingerprint": "f51fc2b5b28c8e33", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 식당 뒷문 앞에 선 채 빤히 시선을 고정한 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The restaurant’s rear doorway opening onto its daylight yard.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Settle the tracking move into a static eye-level wide shot directly opposite the restaurant back door, holding 강영란(흰 머리 여자)'s full body centered within the opening. The frontal symmetry is purposeful here: her planted figure and unwavering look block the visual route back into the restaurant and confront 김지후(강수리영)'s position in the yard.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground; restaurant back doorway in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: restaurant back doorway (open) — The open entrance faces the yard and camera directly; used as Frames her complete figure and fixes the confrontation axis; restaurant yard (visible in daytime); used as Provides the exterior space separating the viewer from the doorway.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 식당 뒷문 앞에 선 채 빤히 시선을 고정한 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The restaurant’s rear doorway opening onto its daylight yard.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Settle the tracking move into a static eye-level wide shot directly opposite the restaurant back door, holding 강영란(흰 머리 여자)'s full body centered within the opening. The frontal symmetry is purposeful here: her planted figure and unwavering look block the visual route back into the restaurant and confront 김지후(강수리영)'s position in the yard.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground; restaurant back doorway in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: restaurant back doorway (open) — The open entrance faces the yard and camera directly; used as Frames her complete figure and fixes the confrontation axis; restaurant yard (visible in daytime); used as Provides the exterior space separating the viewer from the doorway.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S37sh6__bgfirst_bg.png", "asset_id": "db2abe11-415c-4eb0-b72b-d4530314941e", "input_asset_ids": [ "b959ff0a-3252-4c4a-b1a7-eae89a279e11", "6570d7b4-ef3a-4d7b-a02f-ba048e7accc7", "69cb1065-4089-4f87-94f0-4059c3fec052" ] }, "S37sh6": { "input_fingerprint": "120856d367bdd87c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 식당 뒷문 앞에 선 채 빤히 시선을 고정한 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The restaurant’s rear doorway opening onto its daylight yard. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her arrowhead-shaped bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 식당 뒷문 앞에 선 채 빤히 시선을 고정한 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The restaurant’s rear doorway opening onto its daylight yard. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her arrowhead-shaped bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 식당 뒷문 앞에 선 채 빤히 시선을 고정한 강영란(흰 머리 여자)의 정면 전신.\n\nLOCATION (lock): The restaurant’s rear doorway opening onto its daylight yard. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her arrowhead-shaped bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S37sh6__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S37sh6.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L31B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_restaurant_compound_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "인물 레퍼런스의 의상(녹색 한복)과 외모를 정확하게 반영했으며, 제시된 로케이션 사진의 공간 질감을 충실히 살려 정면 전신 샷을 훌륭하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "로케이션 사진의 디테일을 변형해 건물 외벽을 임의로 칠했으며, 무엇보다 캐릭터 레퍼런스의 핵심인 의상(녹색 한복)을 전혀 다른 옷으로 대체하여 감점되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "인물 레퍼런스의 의상(녹색 한복)과 외모를 정확하게 반영했으며, 제시된 로케이션 사진의 공간 질감을 충실히 살려 정면 전신 샷을 훌륭하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "로케이션 사진의 디테일을 변형해 건물 외벽을 임의로 칠했으며, 무엇보다 캐릭터 레퍼런스의 핵심인 의상(녹색 한복)을 전혀 다른 옷으로 대체하여 감점되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물의 전신 프레이밍과 의상이 지시문 및 참조와 훌륭하게 일치하나, 건물의 외관 참조(크림색 벽 등)를 반영하지 못하고 원본 위치 사진의 벽돌을 그대로 유지한 점이 아쉽습니다." }, { "label": "B", "score": 3, "verdict_ko": "프레이밍이 지나치게 넓어 피사체가 작고 캐릭터의 의상이 참조와 전혀 다르며, 위치 참조에 없는 산맥을 배경에 임의로 추가하여 크게 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "인물의 전신 프레이밍과 의상이 지시문 및 참조와 훌륭하게 일치하나, 건물의 외관 참조(크림색 벽 등)를 반영하지 못하고 원본 위치 사진의 벽돌을 그대로 유지한 점이 아쉽습니다." }, { "label": "A", "score": 3, "verdict_ko": "프레이밍이 지나치게 넓어 피사체가 작고 캐릭터의 의상이 참조와 전혀 다르며, 위치 참조에 없는 산맥을 배경에 임의로 추가하여 크게 감점되었습니다." } ] }, "combined": { "totals": { "A": 7, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 7, "B": 15 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "인물 레퍼런스의 의상(녹색 한복)과 외모를 정확하게 반영했으며, 제시된 로케이션 사진의 공간 질감을 충실히 살려 정면 전신 샷을 훌륭하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "로케이션 사진의 디테일을 변형해 건물 외벽을 임의로 칠했으며, 무엇보다 캐릭터 레퍼런스의 핵심인 의상(녹색 한복)을 전혀 다른 옷으로 대체하여 감점되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L31B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_restaurant_compound_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S37sh6__bgfirst_bg.png", "bg_asset_id": "db2abe11-415c-4eb0-b72b-d4530314941e", "bg_record_key": "S37sh6::bgfirst_bg", "chain_winner": false, "authority": "plate", "seed_attached": true }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S38sh3::bgfirst_bg": { "input_fingerprint": "954d2d5fa09936f2", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 고개를 돌린 채 김지후(강수리영)를 서늘하게 쏘아보는 황봉업(황 씨·민숙 집의 거구 남자)의 적대적인 얼굴.\n\nLOCATION (lock): Beside a moored fishing boat along the island pier in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track to 황봉업's eye level just beyond his working reach and settle slightly off his frontal axis in a close three-quarter view. His shoulder and paused hands remain along the lower frame edge while his head turns toward 김지후(강수리영), allowing the hostile eyeline to cut sharply out of frame.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: rope decoration with paper figures (being attached to the fishing boat); used as Appears near the lower edge as evidence of the work he has interrupted; moored fishing boat (moored) — A partial side of the boat is visible behind 황봉업; used as Provides a limited scale reference behind his close profile.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 고개를 돌린 채 김지후(강수리영)를 서늘하게 쏘아보는 황봉업(황 씨·민숙 집의 거구 남자)의 적대적인 얼굴.\n\nLOCATION (lock): Beside a moored fishing boat along the island pier in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track to 황봉업's eye level just beyond his working reach and settle slightly off his frontal axis in a close three-quarter view. His shoulder and paused hands remain along the lower frame edge while his head turns toward 김지후(강수리영), allowing the hostile eyeline to cut sharply out of frame.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: rope decoration with paper figures (being attached to the fishing boat); used as Appears near the lower edge as evidence of the work he has interrupted; moored fishing boat (moored) — A partial side of the boat is visible behind 황봉업; used as Provides a limited scale reference behind his close profile.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S38sh3__bgfirst_bg.png", "asset_id": "be740c75-ec52-4dda-941b-4889ae4a211b", "input_asset_ids": [ "5ae19279-84b9-4729-b7a0-c9fdcd405841", "e98ff21f-0e6d-4176-9c08-156f97437b74" ] }, "S38sh3": { "input_fingerprint": "edd8b89a1c57b4b0", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 고개를 돌린 채 김지후(강수리영)를 서늘하게 쏘아보는 황봉업(황 씨·민숙 집의 거구 남자)의 적대적인 얼굴.\n\nLOCATION (lock): Beside a moored fishing boat along the island pier in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 고개를 돌린 채 김지후(강수리영)를 서늘하게 쏘아보는 황봉업(황 씨·민숙 집의 거구 남자)의 적대적인 얼굴.\n\nLOCATION (lock): Beside a moored fishing boat along the island pier in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 고개를 돌린 채 김지후(강수리영)를 서늘하게 쏘아보는 황봉업(황 씨·민숙 집의 거구 남자)의 적대적인 얼굴.\n\nLOCATION (lock): Beside a moored fishing boat along the island pier in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S38sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S38sh3.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L19B02.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "허용된 유일한 인물인 황봉업만을 프레임에 담아냈으며, 고개를 돌려 적대적으로 쏘아보는 표정과 레퍼런스의 복장 및 배경을 매우 충실히 재현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 허용 인물 목록(황봉업) 외의 인물이 프레임에 포함되었고, 남자가 뒤에 있는 인물을 쏘아보지 않아 텍스트의 핵심 행동과 어긋납니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "허용된 유일한 인물인 황봉업만을 프레임에 담아냈으며, 고개를 돌려 적대적으로 쏘아보는 표정과 레퍼런스의 복장 및 배경을 매우 충실히 재현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 허용 인물 목록(황봉업) 외의 인물이 프레임에 포함되었고, 남자가 뒤에 있는 인물을 쏘아보지 않아 텍스트의 핵심 행동과 어긋납니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "고개를 돌린 채 서늘하게 쏘아보는 적대적인 표정을 잘 연출했으며, 레퍼런스 이미지의 의상(회색 재킷)을 정확히 반영했습니다." }, { "label": "B", "score": 4, "verdict_ko": "고개를 돌린 동작은 묘사되었으나, 지정된 회색 재킷이 아닌 전혀 다른 의상을 입고 있어 복장 일관성에서 크게 벗어났습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "고개를 돌린 채 서늘하게 쏘아보는 적대적인 표정을 잘 연출했으며, 레퍼런스 이미지의 의상(회색 재킷)을 정확히 반영했습니다." }, { "label": "A", "score": 4, "verdict_ko": "고개를 돌린 동작은 묘사되었으나, 지정된 회색 재킷이 아닌 전혀 다른 의상을 입고 있어 복장 일관성에서 크게 벗어났습니다." } ] }, "combined": { "totals": { "A": 8, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 8, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "허용된 유일한 인물인 황봉업만을 프레임에 담아냈으며, 고개를 돌려 적대적으로 쏘아보는 표정과 레퍼런스의 복장 및 배경을 매우 충실히 재현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 허용 인물 목록(황봉업) 외의 인물이 프레임에 포함되었고, 남자가 뒤에 있는 인물을 쏘아보지 않아 텍스트의 핵심 행동과 어긋납니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L19B02.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S38sh3__bgfirst_bg.png", "bg_asset_id": "be740c75-ec52-4dda-941b-4889ae4a211b", "bg_record_key": "S38sh3::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S38sh8": { "input_fingerprint": "c9b38a7a83d8ca44", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)를 향해 친근하고 환한 미소를 짓는 마을 청년#1의 다정한 얼굴.\n\nLOCATION (lock): The open island pier near the decorated fishing boat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dock paving, moored boat, colorful village backdrop, and daylight. Exclude the original welcoming villagers and show the familiar youth approaching with a friendly smile.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 6, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "인물 레퍼런스의 의상(비니의 색상, 로고, 형태)을 가장 정확하게 재현했으며, 지시된 단독 인물의 표정을 잘 담아냈습니다." }, { "label": "C", "score": 6, "verdict_ko": "표정과 배경 구현은 우수하나, 착용한 비니의 색상과 텍스처가 레퍼런스와 일치하지 않습니다." }, { "label": "A", "score": 3, "verdict_ko": "허용된 인물 외에 다른 사람의 뒷모습을 화면에 포함해 엄격한 인물 제한 지시를 위반했으며, 캐릭터의 비니도 누락되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S38sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S38sh3" } }, "groupbg::섬마을_골목": { "input_fingerprint": "9b4c2c04ecb9e5b9", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "섬마을_골목", "tags": [ "S22sh2", "S39sh10", "S39sh3", "S39sh7" ] }, "context_sig": "d3a4ba25c5b94c5a" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A narrow lane through the island’s coastal village, beside homes growing aromatic plants.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 해안 마을: 50여 가구 미만의 알록달록한 지붕을 가진 소담스러운 집들이 모인 마을. 담장에는 새끼줄이 걸려 있다. (특징: 다양한 색의 지붕들; 얕은 담장과 대문; 대문에 걸린 종이 인형과 새끼줄 장식; 집집마다 널린 녹색 석창포 풀)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 일각 - 낮/ 밤\n- 마을 곳곳에 생생하게 폈던 꽃들도 사르륵 사그라져 시들어버린다.\n- 실외. 금월도 해안 마을 일각 - 낮\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A narrow lane through the island’s coastal village, beside homes growing aromatic plants.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 해안 마을: 50여 가구 미만의 알록달록한 지붕을 가진 소담스러운 집들이 모인 마을. 담장에는 새끼줄이 걸려 있다. (특징: 다양한 색의 지붕들; 얕은 담장과 대문; 대문에 걸린 종이 인형과 새끼줄 장식; 집집마다 널린 녹색 석창포 풀)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 일각 - 낮/ 밤\n- 마을 곳곳에 생생하게 폈던 꽃들도 사르륵 사그라져 시들어버린다.\n- 실외. 금월도 해안 마을 일각 - 낮\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_섬마을_골목_f099b6.png", "asset_id": "3c465a81-56fc-4272-bf25-fa1d6f31e771", "input_asset_ids": [ "83c3144a-6163-4d66-b599-2a3d15bdfab2" ], "origin_tag": "S39sh3", "place_text": "A narrow lane through the island’s coastal village, beside homes growing aromatic plants.", "origin_inputs": { "place_text": "A narrow lane through the island’s coastal village, beside homes growing aromatic plants.", "time_of_day_en": "day", "conti_asset_id": "83c3144a-6163-4d66-b599-2a3d15bdfab2" } }, "S39sh3::bgfirst_bg": { "input_fingerprint": "7aba87b75839934b", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 코를 찌르는 악취에 미간을 잔뜩 찌푸리며 코를 쥔 김지후(강수리영)의 굳은 얼굴.\n\nLOCATION (lock): A narrow lane through the island’s coastal village, beside homes growing aromatic plants.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in at 김지후(강수리영)'s eye height from a three-quarter frontal position, stopping in a close frame that contains her clenched expression and the hand pinching her nose. Catch her between steps with the village plants and low wall compressed behind her, making the bodily reaction read before its source is revealed.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: low wall (visible) — Its lane-facing side runs laterally behind 김지후(강수리영); used as Provides partial village context behind her close reaction; green seokchangpo plants (grown in abundance by the houses); used as Appears behind her as the residents' stated explanation for the smell.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 코를 찌르는 악취에 미간을 잔뜩 찌푸리며 코를 쥔 김지후(강수리영)의 굳은 얼굴.\n\nLOCATION (lock): A narrow lane through the island’s coastal village, beside homes growing aromatic plants.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in at 김지후(강수리영)'s eye height from a three-quarter frontal position, stopping in a close frame that contains her clenched expression and the hand pinching her nose. Catch her between steps with the village plants and low wall compressed behind her, making the bodily reaction read before its source is revealed.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: low wall (visible) — Its lane-facing side runs laterally behind 김지후(강수리영); used as Provides partial village context behind her close reaction; green seokchangpo plants (grown in abundance by the houses); used as Appears behind her as the residents' stated explanation for the smell.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S39sh3__bgfirst_bg.png", "asset_id": "0d1e59ad-8d2a-4839-ba88-89bcaf213fc5", "input_asset_ids": [ "83c3144a-6163-4d66-b599-2a3d15bdfab2", "3c465a81-56fc-4272-bf25-fa1d6f31e771" ] }, "S39sh3": { "input_fingerprint": "8c3d5bc97a66de56", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 코를 찌르는 악취에 미간을 잔뜩 찌푸리며 코를 쥔 김지후(강수리영)의 굳은 얼굴.\n\nLOCATION (lock): A narrow lane through the island’s coastal village, beside homes growing aromatic plants. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 코를 찌르는 악취에 미간을 잔뜩 찌푸리며 코를 쥔 김지후(강수리영)의 굳은 얼굴.\n\nLOCATION (lock): A narrow lane through the island’s coastal village, beside homes growing aromatic plants. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 코를 찌르는 악취에 미간을 잔뜩 찌푸리며 코를 쥔 김지후(강수리영)의 굳은 얼굴.\n\nLOCATION (lock): A narrow lane through the island’s coastal village, beside homes growing aromatic plants. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S39sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S39sh3.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_섬마을_골목_f099b6.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 찡그린 표정과 코를 쥔 동작, 장소는 잘 구현했으나 레퍼런스의 의상과 일치하지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "의상은 레퍼런스와 일치하나 인물 뒤에 여분의 머리가 생성된 치명적인 신체 구조 오류(Hard Violation)가 발생했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 찡그린 표정과 코를 쥔 동작, 장소는 잘 구현했으나 레퍼런스의 의상과 일치하지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "의상은 레퍼런스와 일치하나 인물 뒤에 여분의 머리가 생성된 치명적인 신체 구조 오류(Hard Violation)가 발생했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "머리 뒤에 정체불명의 머리카락 형태가 중복 생성되는 심각한 신체 오류가 발생했으며, 뼈 목걸이가 여전히 존재합니다." }, { "label": "B", "score": 7, "verdict_ko": "코를 쥐고 미간을 찌푸린 표정 연기를 사실적으로 잘 구현했으며, 목걸이를 제거하라는 지시와 지정된 배경도 정확히 따랐습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 2, "verdict_ko": "머리 뒤에 정체불명의 머리카락 형태가 중복 생성되는 심각한 신체 오류가 발생했으며, 뼈 목걸이가 여전히 존재합니다." }, { "label": "A", "score": 7, "verdict_ko": "코를 쥐고 미간을 찌푸린 표정 연기를 사실적으로 잘 구현했으며, 목걸이를 제거하라는 지시와 지정된 배경도 정확히 따랐습니다." } ] }, "combined": { "totals": { "A": 14, "B": 5 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 5 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 찡그린 표정과 코를 쥔 동작, 장소는 잘 구현했으나 레퍼런스의 의상과 일치하지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "의상은 레퍼런스와 일치하나 인물 뒤에 여분의 머리가 생성된 치명적인 신체 구조 오류(Hard Violation)가 발생했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_섬마을_골목_f099b6.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S39sh3__bgfirst_bg.png", "bg_asset_id": "0d1e59ad-8d2a-4839-ba88-89bcaf213fc5", "bg_record_key": "S39sh3::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "섬마을_골목", "groupbg_asset_id": "3c465a81-56fc-4272-bf25-fa1d6f31e771" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S39sh7": { "input_fingerprint": "a6c536d15afe5500", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 야트막한 담장 너머로 일제히 고개를 내밀고 김지후(강수리영)를 빤히 쳐다보는 수많은 한국인 마을 사람들의 기묘한 시선.\n\nLOCATION (lock): A residential village lane bordered by low walls and house yards. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The houses continue to display large quantities of green seokchangpo plants, and Jihu remains without her necklace.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "레퍼런스의 특징적인 돌담과 식물(석창포)을 가장 정확하게 유지하면서 마을 사람들의 시선을 무난하게 연출했습니다." }, { "label": "B", "score": 3, "verdict_ko": "담장 너머로 고개를 내미는 행동은 잘 표현되었으나, 좌측 돌담이 시멘트 블록으로 바뀌는 심각한 장소 오류가 있고 주인공이 너무 작게 묘사되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "구도와 인물의 행동 연출은 훌륭하지만, B와 마찬가지로 레퍼런스의 핵심인 돌담을 시멘트 블록으로 왜곡하여 장소 일관성을 훼손했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S39sh3_sel.png" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S39sh3" }, "lane_policy": "ab_select_bypass:prev" }, "S39sh10": { "input_fingerprint": "bf31f8a10515ffeb", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 숲 쪽을 향해 손가락을 뻗은 채 단호한 눈빛으로 입을 벌리고 있는 한국인 마을 주민의 진지한 얼굴.\n\nLOCATION (lock): The village lane at the side of a low-walled coastal home, facing toward the forest. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same low wall, clustered villagers, village houses, and bright daylight. Exclude the collective staring pose and focus on one resident warning toward the forest.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The seokchangpo plants remain cultivated at the village houses, and Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 5, "B": 8, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "이전 샷에 확립된 군중의 배치와 복장 연속성이 떨어지며, 인물의 외형이 다소 변형되었습니다." }, { "label": "B", "score": 8, "verdict_ko": "이전 샷의 군중 위치와 복장을 완벽히 유지하면서 텍스트가 요구한 경고 동작을 정확히 연출했고, 지후의 외모도 레퍼런스와 일치합니다." }, { "label": "C", "score": 4, "verdict_ko": "이전 샷에 없던 새로운 인물이 전면에 등장했으며, 기존 주민들의 위치와 연속성이 크게 훼손되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S39sh7_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S39sh7" } }, "groupbg::숲_입구": { "input_fingerprint": "1bdb5c55dbd60b32", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "숲_입구", "tags": [ "S40sh1", "S40sh3" ] }, "context_sig": "41319975e9383340" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: The shaded mouth of a narrow forest trail at the edge of the village in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 숲속 입구: 밝은 마을 쪽에서 이어지는 숲의 시작점. 숲 안쪽은 어둡다. (특징: 경계가 뚜렷한 빛과 어둠; 좁은 오솔길의 시작 지점; 우거진 해송과 수풀)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 숲속 입구 - 낮\n- 수리영이 숲으로 통하는 오솔길 입구에 서 있다.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: The shaded mouth of a narrow forest trail at the edge of the village in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 숲속 입구: 밝은 마을 쪽에서 이어지는 숲의 시작점. 숲 안쪽은 어둡다. (특징: 경계가 뚜렷한 빛과 어둠; 좁은 오솔길의 시작 지점; 우거진 해송과 수풀)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 숲속 입구 - 낮\n- 수리영이 숲으로 통하는 오솔길 입구에 서 있다.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_숲_입구_b563a8.png", "asset_id": "0482c48b-e76f-4e55-a34a-e95f89d77ed9", "input_asset_ids": [ "f64abac1-e735-44a0-82d8-fc97acf653a9" ], "origin_tag": "S40sh1", "place_text": "The shaded mouth of a narrow forest trail at the edge of the village in daylight.", "origin_inputs": { "place_text": "The shaded mouth of a narrow forest trail at the edge of the village in daylight.", "time_of_day_en": "day", "conti_asset_id": "f64abac1-e735-44a0-82d8-fc97acf653a9" } }, "S40sh1::bgfirst_bg": { "input_fingerprint": "c4269c2692004e6e", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한낮임에도 빛이 들지 않아 시커먼 오솔길 입구에 우두커니 선 김지후(강수리영)의 전신.\n\nLOCATION (lock): The shaded mouth of a narrow forest trail at the edge of the village in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold a static wide shot from outside the path entrance at eye level, with 김지후(강수리영)'s full figure placed low and near center against the opening. Show her from behind with her weight arrested at the threshold, while the dark path continues above and beyond her into the forest, preserving the axis for the coming forward move.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 김지후(강수리영) in the lower-center of the frame, midground, looks toward dark forest path; dark forest path in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: forest path entrance (open) — The entrance faces the camera, opening directly into the path's depth; used as Frames 김지후(강수리영)'s solitary full figure at the decision point; forest path (dark despite daytime); used as Extends behind her as the destination of her fixed attention; forest (light does not appear to reach its interior); used as Encloses the path and isolates her silhouette within the entrance.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한낮임에도 빛이 들지 않아 시커먼 오솔길 입구에 우두커니 선 김지후(강수리영)의 전신.\n\nLOCATION (lock): The shaded mouth of a narrow forest trail at the edge of the village in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold a static wide shot from outside the path entrance at eye level, with 김지후(강수리영)'s full figure placed low and near center against the opening. Show her from behind with her weight arrested at the threshold, while the dark path continues above and beyond her into the forest, preserving the axis for the coming forward move.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 김지후(강수리영) in the lower-center of the frame, midground, looks toward dark forest path; dark forest path in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: forest path entrance (open) — The entrance faces the camera, opening directly into the path's depth; used as Frames 김지후(강수리영)'s solitary full figure at the decision point; forest path (dark despite daytime); used as Extends behind her as the destination of her fixed attention; forest (light does not appear to reach its interior); used as Encloses the path and isolates her silhouette within the entrance.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S40sh1__bgfirst_bg.png", "asset_id": "6c23b675-70ee-4cf5-9828-3afbefec70c1", "input_asset_ids": [ "f64abac1-e735-44a0-82d8-fc97acf653a9", "0482c48b-e76f-4e55-a34a-e95f89d77ed9" ] }, "S40sh1": { "input_fingerprint": "e2940e12b71c35e9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한낮임에도 빛이 들지 않아 시커먼 오솔길 입구에 우두커니 선 김지후(강수리영)의 전신.\n\nLOCATION (lock): The shaded mouth of a narrow forest trail at the edge of the village in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace at the forest entrance.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한낮임에도 빛이 들지 않아 시커먼 오솔길 입구에 우두커니 선 김지후(강수리영)의 전신.\n\nLOCATION (lock): The shaded mouth of a narrow forest trail at the edge of the village in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace at the forest entrance.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한낮임에도 빛이 들지 않아 시커먼 오솔길 입구에 우두커니 선 김지후(강수리영)의 전신.\n\nLOCATION (lock): The shaded mouth of a narrow forest trail at the edge of the village in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace at the forest entrance.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S40sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S40sh1.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_숲_입구_b563a8.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "캐릭터가 뒤를 돌아보고 있어 인물의 신원(얼굴)과 지시된 목걸이 부재 상태를 확인할 수 없습니다." }, { "label": "B", "score": 10, "verdict_ko": "어두운 오솔길 입구에 선 인물의 전신을 정확한 프레이밍으로 담아냈으며, 얼굴 일치도와 목걸이가 없는 의상 상태까지 완벽하게 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "캐릭터가 뒤를 돌아보고 있어 인물의 신원(얼굴)과 지시된 목걸이 부재 상태를 확인할 수 없습니다." }, { "label": "B", "score": 10, "verdict_ko": "어두운 오솔길 입구에 선 인물의 전신을 정확한 프레이밍으로 담아냈으며, 얼굴 일치도와 목걸이가 없는 의상 상태까지 완벽하게 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 배경에서 뼈 목걸이를 뺀 상태로 우두커니 서 있는 캐릭터의 전신을 완벽하게 구현했습니다." }, { "label": "B", "score": 8, "verdict_ko": "배경과 전신 구도는 맞으나, 캐릭터가 뒤를 돌고 있어 얼굴과 표정('우두커니 선' 모습)을 온전히 확인할 수 없습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "지정된 배경에서 뼈 목걸이를 뺀 상태로 우두커니 서 있는 캐릭터의 전신을 완벽하게 구현했습니다." }, { "label": "A", "score": 8, "verdict_ko": "배경과 전신 구도는 맞으나, 캐릭터가 뒤를 돌고 있어 얼굴과 표정('우두커니 선' 모습)을 온전히 확인할 수 없습니다." } ] }, "combined": { "totals": { "A": 14, "B": 20 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 20 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "캐릭터가 뒤를 돌아보고 있어 인물의 신원(얼굴)과 지시된 목걸이 부재 상태를 확인할 수 없습니다." }, { "label": "B", "score": 10, "verdict_ko": "어두운 오솔길 입구에 선 인물의 전신을 정확한 프레이밍으로 담아냈으며, 얼굴 일치도와 목걸이가 없는 의상 상태까지 완벽하게 구현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_숲_입구_b563a8.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S40sh1__bgfirst_bg.png", "bg_asset_id": "6c23b675-70ee-4cf5-9828-3afbefec70c1", "bg_record_key": "S40sh1::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "숲_입구", "groupbg_asset_id": "0482c48b-e76f-4e55-a34a-e95f89d77ed9" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S40sh3": { "input_fingerprint": "222723b1af39aac9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두운 숲속에서 밖의 수리영을 향해 시선을 고정하고 선 검은 염소의 기묘한 자태.\n\nLOCATION (lock): Just inside the dark tree line beyond the village-side forest entrance. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace as the black goat watches her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 5, "B": 7, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "장소의 형태학적 일치도는 매우 뛰어나나, 염소가 프레임 내의 인물이 아닌 카메라를 바라보고 있어 '수리영을 향해 시선을 고정'한다는 핵심 지시를 위반했습니다." }, { "label": "B", "score": 7, "verdict_ko": "인물을 배제하라는 네거티브 지시를 정확히 따랐으며, 카메라를 향한 염소의 시선을 통해 프레임 밖의 대상을 바라보는 설정을 가장 충실하게 구현했습니다." }, { "label": "C", "score": 4, "verdict_ko": "인물이 뒤돌아 걸어가고 있어 이전 샷의 상태 유지 지시를 위반했으며, 염소의 시선 방향도 텍스트의 맥락과 일치하지 않습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S40sh1_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S40sh1" }, "lane_policy": "share_plan_prev_bgonly" }, "S41sh2": { "input_fingerprint": "0e8710c35769983b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 이끼가 잔뜩 낀 허리 높이의 비석들 앞마다 붉게 썩은 육고기 덩어리들이 놓여 있는 기괴한 풍경.\n\nLOCATION (lock): A deep-forest burial-marker area with mossy waist-high stones and rotting meat offerings. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Each moss-covered stone marker retains its offering of rotting meat, and Jihu remains without her necklace.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "A": 7, "C": 6 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "이끼 낀 비석들 앞마다 붉게 썩은 고깃덩어리가 놓인 기괴한 풍경을 매우 사실적이고 지시문에 부합하게 묘사했습니다." }, { "label": "A", "score": 7, "verdict_ko": "지시된 풍경을 잘 구현했으나, 고깃덩어리의 형태가 특정 장기(심장)에 너무 가깝게 묘사되었습니다." }, { "label": "C", "score": 6, "verdict_ko": "비석과 숲의 묘사는 좋으나, 고기가 '붉게 썩은' 느낌보다는 검게 그을리거나 변색된 형태로 표현되어 아쉽습니다." } ], "refs": [], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S41sh8": { "input_fingerprint": "443db3a6f3d0f20f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 멀찍이 떨어진 굵은 나무 기둥 너머로 절반쯤 가려진 채 허공에 뻗어 있는 뾰족한 짐승 손.\n\nLOCATION (lock): Behind a thick pine trunk overlooking the forest marker path. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace while the hidden clawed creature stalks her among the trees.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 8, "B": 5, "A": 4 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "카메라 위치를 전경의 나무 뒤로 잡고 짐승 손이 '멀찍이 떨어진' 원경의 나무에서 나오도록 한 공간감과 '절반쯤 가려진' 연출을 유일하게 정확히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "뾰족한 짐승 손의 묘사는 좋으나, '멀찍이 떨어진'이라는 지시를 어기고 카메라 바로 앞 근경의 나무에서 손이 등장해 Staging에 실패했습니다." }, { "label": "A", "score": 4, "verdict_ko": "'멀찍이 떨어진' 원경 연출을 무시하고 손을 근접하게 배치했으며, 짐승 손 특유의 날카로운 묘사도 상대적으로 부족합니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S41sh2_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S41sh2" }, "lane_policy": "share_plan_prev_bgonly" }, "groupbg::국도변_주유소": { "input_fingerprint": "2baff0bc5d79ba8a", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "국도변_주유소", "tags": [ "S42sh2" ] }, "context_sig": "2802e8542dd44564" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: Beside a fuel pump at a rural roadside gas station in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n국도변 주유소: 지방 도로 옆에 위치한 한산한 형태의 주유 공간. (특징: 주유기 구조물; 콘크리트 포장 바닥; 밝은 캐노피 조명)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 국도변 주유소 - 낮\n- 주유기 옆에 서 있는 아반테.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: Beside a fuel pump at a rural roadside gas station in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n국도변 주유소: 지방 도로 옆에 위치한 한산한 형태의 주유 공간. (특징: 주유기 구조물; 콘크리트 포장 바닥; 밝은 캐노피 조명)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 국도변 주유소 - 낮\n- 주유기 옆에 서 있는 아반테.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_국도변_주유소_e533a2.png", "asset_id": "66db7c6a-a2fb-4a62-ae67-9d339b7f3e2c", "input_asset_ids": [ "f0f5eaae-b59e-4dd9-9161-735a69db1c08" ], "origin_tag": "S42sh2", "place_text": "Beside a fuel pump at a rural roadside gas station in daylight.", "origin_inputs": { "place_text": "Beside a fuel pump at a rural roadside gas station in daylight.", "time_of_day_en": "day", "conti_asset_id": "f0f5eaae-b59e-4dd9-9161-735a69db1c08" } }, "S42sh2::bgfirst_bg": { "input_fingerprint": "aae58afd8a24ea64", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 정인우의 손에 들린 핸드폰 화면 위로 여러 섬에서 멀리 떨어진 '금월도'라는 위치가 선명하게 뜬 찰나.\n\nLOCATION (lock): Beside a fuel pump at a rural roadside gas station in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The static camera looks almost perpendicularly down onto the phone, keeping the screen within the central two-fifths while 정인우's supporting fingers remain at the edges as scale reference. The isolated island and the label “금월도” are sharp at center, with several neighboring islands retained around it to prove the geographic separation.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: map phone screen in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 핸드폰 지도앱 화면 (지도앱이 열린 상태) — The illuminated front face is visible, showing “금월도” separated from several nearby islands; used as Primary information surface held nearly square to the lens.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 정인우의 손에 들린 핸드폰 화면 위로 여러 섬에서 멀리 떨어진 '금월도'라는 위치가 선명하게 뜬 찰나.\n\nLOCATION (lock): Beside a fuel pump at a rural roadside gas station in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The static camera looks almost perpendicularly down onto the phone, keeping the screen within the central two-fifths while 정인우's supporting fingers remain at the edges as scale reference. The isolated island and the label “금월도” are sharp at center, with several neighboring islands retained around it to prove the geographic separation.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: map phone screen in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 핸드폰 지도앱 화면 (지도앱이 열린 상태) — The illuminated front face is visible, showing “금월도” separated from several nearby islands; used as Primary information surface held nearly square to the lens.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S42sh2__bgfirst_bg.png", "asset_id": "a87264e1-b809-4bb8-b5d5-fd7438eabec3", "input_asset_ids": [ "f0f5eaae-b59e-4dd9-9161-735a69db1c08", "66db7c6a-a2fb-4a62-ae67-9d339b7f3e2c" ] }, "S42sh2": { "input_fingerprint": "21c7b0934bdb1f6a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 정인우의 손에 들린 핸드폰 화면 위로 여러 섬에서 멀리 떨어진 '금월도'라는 위치가 선명하게 뜬 찰나.\n\nLOCATION (lock): Beside a fuel pump at a rural roadside gas station in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to have one arm immobilized in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 정인우의 손에 들린 핸드폰 화면 위로 여러 섬에서 멀리 떨어진 '금월도'라는 위치가 선명하게 뜬 찰나.\n\nLOCATION (lock): Beside a fuel pump at a rural roadside gas station in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to have one arm immobilized in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 정인우의 손에 들린 핸드폰 화면 위로 여러 섬에서 멀리 떨어진 '금월도'라는 위치가 선명하게 뜬 찰나.\n\nLOCATION (lock): Beside a fuel pump at a rural roadside gas station in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to have one arm immobilized in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S42sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S42sh2.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_국도변_주유소_e533a2.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 '금월도' 텍스트와 지도 화면이 선명하게 클로즈업되어 샷 지문을 완벽하게 구현했으며, 깁스 설정도 잘 반영되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 얼굴과 배경은 잘 묘사되었으나, 가장 중요한 핸드폰 화면의 '금월도' 텍스트가 오타로 출력되어 지문 충실도가 떨어집니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 '금월도' 텍스트와 지도 화면이 선명하게 클로즈업되어 샷 지문을 완벽하게 구현했으며, 깁스 설정도 잘 반영되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 얼굴과 배경은 잘 묘사되었으나, 가장 중요한 핸드폰 화면의 '금월도' 텍스트가 오타로 출력되어 지문 충실도가 떨어집니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "레퍼런스와 동일한 주유소 건축물과 주유기를 완벽하게 재현했으며, 인물의 외모와 깁스 설정, 화면 내 지도 연출까지 지시사항을 충실히 따랐습니다." }, { "label": "B", "score": 4, "verdict_ko": "화면 내 '금월도' 텍스트와 1인칭 시점 연출은 훌륭하나, 주유기의 형태와 색상이 레퍼런스와 완전히 달라 로케이션 고정(Location lock) 조건을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "레퍼런스와 동일한 주유소 건축물과 주유기를 완벽하게 재현했으며, 인물의 외모와 깁스 설정, 화면 내 지도 연출까지 지시사항을 충실히 따랐습니다." }, { "label": "A", "score": 4, "verdict_ko": "화면 내 '금월도' 텍스트와 1인칭 시점 연출은 훌륭하나, 주유기의 형태와 색상이 레퍼런스와 완전히 달라 로케이션 고정(Location lock) 조건을 위반했습니다." } ] }, "combined": { "totals": { "A": 11, "B": 11 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 11 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 '금월도' 텍스트와 지도 화면이 선명하게 클로즈업되어 샷 지문을 완벽하게 구현했으며, 깁스 설정도 잘 반영되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 얼굴과 배경은 잘 묘사되었으나, 가장 중요한 핸드폰 화면의 '금월도' 텍스트가 오타로 출력되어 지문 충실도가 떨어집니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_국도변_주유소_e533a2.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S42sh2__bgfirst_bg.png", "bg_asset_id": "a87264e1-b809-4bb8-b5d5-fd7438eabec3", "bg_record_key": "S42sh2::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "국도변_주유소", "groupbg_asset_id": "66db7c6a-a2fb-4a62-ae67-9d339b7f3e2c" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S43sh2": { "input_fingerprint": "9743c5d6387d11f4", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 이끼 낀 비석과 붉은 썩은 고기 앞에서 눈을 휘둥그레 뜬 채 우뚝 선 김지후(강수리영)의 전신.\n\nLOCATION (lock): The looping forest path directly before a moss-covered marker and its rotting meat offering. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same mossy waist-high stones, rotten red meat, pine trunks, and subdued daylight. Exclude the hidden claw and show the woman realizing she has returned to the same place.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The mossy marker and its rotting-meat offering remain in exactly the same arrangement as before; Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 4, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "'전신' 샷과 '눈을 휘둥그레 뜬' 표정을 완벽히 구현했으며, 이전 샷의 허리 높이 비석 스케일과 공간 배치를 물리적 오류 없이 유지했습니다." }, { "label": "A", "score": 4, "verdict_ko": "요구된 샷과 표정은 반영되었으나, 인물과 비석의 스케일이 심하게 왜곡되어 레퍼런스의 물리적 환경을 파괴한 중대한 오류가 있습니다." }, { "label": "B", "score": 4, "verdict_ko": "A와 동일하게 비석이 발목 높이로 축소되는 물리적 비율 왜곡이 발생하여 이전 샷의 환경 연속성을 지키지 못했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S41sh8_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 비석 앞 썩은 고기 공물: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S41sh8" } }, "S43sh4": { "input_fingerprint": "80aa9e1c10bc0b72", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 이끼 낀 바위 표면에 '수 리 영'이라는 세 글자가 선명하게 음각으로 새겨져 있는 비석의 클로즈업.\n\nLOCATION (lock): At the face of a mossy stone marker within the deep island forest. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The name “Su Ri-yeong” remains carved into the moss-covered marker, with the rotting meat still placed before it.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 6, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "요구된 클로즈업 샷과 세 글자의 음각을 정확히 구현했으며, 레퍼런스의 둥근 비석 형태와 고기를 잘 유지했습니다." }, { "label": "A", "score": 6, "verdict_ko": "텍스트 각인과 클로즈업 프레이밍은 훌륭하나, 비석의 상단이 레퍼런스의 둥근 형태와 달리 평평하게 묘사되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "비석에 프롬프트가 요구하지 않은 '1670'이라는 숫자를 임의로 추가하여 텍스트 제약을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S43sh2_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S43sh2" }, "lane_policy": "share_plan_prev_bgonly" }, "S44sh1::bgfirst_bg": { "input_fingerprint": "289c13c17c2e19dc", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한낮의 햇살이 스며드는 도축장 안, 핏자국 가득한 바닥에 납작 엎드려 바짝 움츠러든 마을 청년#1과 한국인 청년 무리의 전신.\n\nLOCATION (lock): The bloodstained main floor inside the slaughterhouse, lit by daylight through a broken window.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height across the room, the static wide frame shows 마을 청년#1 and the other Korean village youths pressed flat along the blood-marked floor, their full bodies forming an uneven receding line toward the far side. The broken window and hanging implements remain environmental evidence rather than foreground spectacle, while the first youth lifts only his eyes toward 고산들 outside the frame.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 깨진 창 (깨진 상태) — The damaged opening faces the camera obliquely, with midday light entering through it; used as Admits the stated midday sunlight and breaks the upper boundary of the room; 바닥의 핏자국 (핏자국이 가득한 상태); used as Provides the low plane beneath the prostrate group and makes their submission physically legible; 도축용 칼과 고기 걸이 금속 걸개 (빛을 받아 반짝이는 상태) — Their edges and hooked profiles catch the incoming light at oblique angles; used as Adds sharp vertical and suspended shapes behind the youths; 벽의 핏자국 (핏자국이 선명한 상태); used as Extends the marked enclosure behind the group.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한낮의 햇살이 스며드는 도축장 안, 핏자국 가득한 바닥에 납작 엎드려 바짝 움츠러든 마을 청년#1과 한국인 청년 무리의 전신.\n\nLOCATION (lock): The bloodstained main floor inside the slaughterhouse, lit by daylight through a broken window.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height across the room, the static wide frame shows 마을 청년#1 and the other Korean village youths pressed flat along the blood-marked floor, their full bodies forming an uneven receding line toward the far side. The broken window and hanging implements remain environmental evidence rather than foreground spectacle, while the first youth lifts only his eyes toward 고산들 outside the frame.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 깨진 창 (깨진 상태) — The damaged opening faces the camera obliquely, with midday light entering through it; used as Admits the stated midday sunlight and breaks the upper boundary of the room; 바닥의 핏자국 (핏자국이 가득한 상태); used as Provides the low plane beneath the prostrate group and makes their submission physically legible; 도축용 칼과 고기 걸이 금속 걸개 (빛을 받아 반짝이는 상태) — Their edges and hooked profiles catch the incoming light at oblique angles; used as Adds sharp vertical and suspended shapes behind the youths; 벽의 핏자국 (핏자국이 선명한 상태); used as Extends the marked enclosure behind the group.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S44sh1__bgfirst_bg.png", "asset_id": "5cdb3105-543b-4acc-b19c-8bf1f741eadd", "input_asset_ids": [ "9dd7a517-67a9-4bc6-aebd-999d26709792", "0cf5113e-e387-4d64-9d8b-6afe9e7e8352" ] }, "S44sh1": { "input_fingerprint": "28390c657eca6977", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한낮의 햇살이 스며드는 도축장 안, 핏자국 가득한 바닥에 납작 엎드려 바짝 움츠러든 마을 청년#1과 한국인 청년 무리의 전신.\n\nLOCATION (lock): The bloodstained main floor inside the slaughterhouse, lit by daylight through a broken window. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The slaughterhouse retains the bloodstains on its floor and walls and the fixed array of slaughter tools.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한낮의 햇살이 스며드는 도축장 안, 핏자국 가득한 바닥에 납작 엎드려 바짝 움츠러든 마을 청년#1과 한국인 청년 무리의 전신.\n\nLOCATION (lock): The bloodstained main floor inside the slaughterhouse, lit by daylight through a broken window. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The slaughterhouse retains the bloodstains on its floor and walls and the fixed array of slaughter tools.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한낮의 햇살이 스며드는 도축장 안, 핏자국 가득한 바닥에 납작 엎드려 바짝 움츠러든 마을 청년#1과 한국인 청년 무리의 전신.\n\nLOCATION (lock): The bloodstained main floor inside the slaughterhouse, lit by daylight through a broken window. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The slaughterhouse retains the bloodstains on its floor and walls and the fixed array of slaughter tools.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S44sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S44sh1.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B01.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "마을 청년#1을 포함한 청년 무리 전체가 '바닥에 납작 엎드려 바짝 움츠러든' 포즈를 지문 그대로 정확하고 충실하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "대다수의 인물들이 바닥에 납작 엎드리지 않고 쪼그려 앉아 있어 핵심 포즈 지시인 '납작 엎드려'를 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "마을 청년#1을 포함한 청년 무리 전체가 '바닥에 납작 엎드려 바짝 움츠러든' 포즈를 지문 그대로 정확하고 충실하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "대다수의 인물들이 바닥에 납작 엎드리지 않고 쪼그려 앉아 있어 핵심 포즈 지시인 '납작 엎드려'를 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "마을 청년#1과 청년 무리 모두가 바닥에 납작 엎드려 있는 자세(납작 엎드려)를 지문에 맞게 정확히 연출했습니다." }, { "label": "A", "score": 4, "verdict_ko": "대부분의 인물이 바닥에 엎드리지 않고 쪼그려 앉아 있어 '납작 엎드려'라는 핵심 행동 지시를 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "마을 청년#1과 청년 무리 모두가 바닥에 납작 엎드려 있는 자세(납작 엎드려)를 지문에 맞게 정확히 연출했습니다." }, { "label": "B", "score": 4, "verdict_ko": "대부분의 인물이 바닥에 엎드리지 않고 쪼그려 앉아 있어 '납작 엎드려'라는 핵심 행동 지시를 위반했습니다." } ] }, "combined": { "totals": { "A": 19, "B": 9 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 19, "B": 9 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "마을 청년#1을 포함한 청년 무리 전체가 '바닥에 납작 엎드려 바짝 움츠러든' 포즈를 지문 그대로 정확하고 충실하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "대다수의 인물들이 바닥에 납작 엎드리지 않고 쪼그려 앉아 있어 핵심 포즈 지시인 '납작 엎드려'를 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B01.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S44sh1__bgfirst_bg.png", "bg_asset_id": "5cdb3105-543b-4acc-b19c-8bf1f741eadd", "bg_record_key": "S44sh1::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S44sh7": { "input_fingerprint": "d0f56c6cdb217143", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 강영란(흰 머리 여자)의 날카로운 손톱이 마을 청년#1의 목줄기를 깊게 파고든 상태로 멈춘 mid-impact 순간.\n\nLOCATION (lock): The slaughterhouse killing floor beside the hanging knives and butchering tools. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same blood-stained floor, hanging tools, broken-window daylight, and cowering youths. Exclude their earlier prostrate arrangement from the foreground and show the elder slashing one youth's throat.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran's elongated claw has opened the youth's throat, and fresh blood remains on her hand as he begins collapsing.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (각성 상태) (60대 후반 한국인 여성, 고운 노년 얼굴, 가지런히 틀어 묶은 머리, 날카로운 얼굴선); 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 5, "B": 3, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "날카로운 손톱이 구현되었고 신체 접촉이 있으나, 목을 깊게 파고드는 대신 가슴 부위에 닿아 있으며 출혈 묘사가 누락되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "청년의 목에 손이 닿아 있으나, 샷 텍스트의 핵심인 날카로운 손톱이 완전히 누락되었고 출혈도 표현되지 않았습니다." }, { "label": "C", "score": 4, "verdict_ko": "날카로운 손톱은 묘사되었으나 청년의 신체에 닿지 않고 허공에 떠 있어 타격 순간의 지시를 충족하지 못했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S44sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자) (각성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S44sh1" } }, "S44sh11": { "input_fingerprint": "31571a9161956614", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 피범벅이 된 자신의 뾰족한 손톱에 혀끝을 댄 채 태연하게 미소 짓는 강영란(흰 머리 여자)의 기괴한 얼굴.\n\nLOCATION (lock): The bloodstained work area inside the daylight-lit slaughterhouse. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same slaughterhouse walls, bright window shafts, blood stains, and metallic tools. Exclude the other youths from the close frame and focus on the elder licking blood from her claw.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The youth lies dying from the slashed throat, while Youngran's elongated fingernails remain smeared with his blood.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (각성 상태) (60대 후반 한국인 여성, 고운 노년 얼굴, 가지런히 틀어 묶은 머리, 날카로운 얼굴선) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "이전 샷의 녹색 한복을 유지하며 피 묻은 손톱을 핥으며 미소 짓는 기괴한 표정을 가장 정확하게 재현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "표정과 행동 연출은 적절하나, 이전 샷에서 연속되어야 할 녹색 한복 의상이 어두운 색상의 다른 옷으로 변경되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "의상과 표정은 지시문과 일치하지만, 손가락의 개수와 구조가 비정상적으로 묘사된 치명적인 해부학적 오류가 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S44sh7_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자) (각성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S44sh7" } }, "S45sh3::bgfirst_bg": { "input_fingerprint": "84b4dc037f0a6662", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 몸을 반쯤 돌린 채 고산들을 날카롭게 쏘아보는 강영란(흰 머리 여자)의 서늘한 얼굴.\n\nLOCATION (lock): A descending maritime-pine trail leading away from the forest slaughterhouse in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 강영란(흰 머리 여자)'s eye height, the dolly-in ends between her frontal and profile lines as she arrests her turn, placing her cold three-quarter face on one side and leaving gaze room toward 고산들 beyond the opposite edge. The close framing preserves part of her turned shoulder and the forest path behind her, so the accusation reads as an interruption of forward movement rather than a posed portrait.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-left of the frame, foreground, looks toward 고산들 beyond the opposite frame edge.\n- KEY BACKGROUND ELEMENTS: 숲길 (보이는 상태); used as Retains the direction of travel behind her turned shoulder; 숲의 나무와 수풀 (보이는 상태); used as Creates compressed vertical depth behind the confrontation close-up.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 몸을 반쯤 돌린 채 고산들을 날카롭게 쏘아보는 강영란(흰 머리 여자)의 서늘한 얼굴.\n\nLOCATION (lock): A descending maritime-pine trail leading away from the forest slaughterhouse in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 강영란(흰 머리 여자)'s eye height, the dolly-in ends between her frontal and profile lines as she arrests her turn, placing her cold three-quarter face on one side and leaving gaze room toward 고산들 beyond the opposite edge. The close framing preserves part of her turned shoulder and the forest path behind her, so the accusation reads as an interruption of forward movement rather than a posed portrait.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-left of the frame, foreground, looks toward 고산들 beyond the opposite frame edge.\n- KEY BACKGROUND ELEMENTS: 숲길 (보이는 상태); used as Retains the direction of travel behind her turned shoulder; 숲의 나무와 수풀 (보이는 상태); used as Creates compressed vertical depth behind the confrontation close-up.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S45sh3__bgfirst_bg.png", "asset_id": "366089ca-3869-4446-afc8-94594ed51e1e", "input_asset_ids": [ "67c0e71f-b039-4941-910a-5a32415c9c01", "dded1895-8314-4902-9d6c-855ef980a51c" ] }, "S45sh3": { "input_fingerprint": "ad57ec37da24948f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 몸을 반쯤 돌린 채 고산들을 날카롭게 쏘아보는 강영란(흰 머리 여자)의 서늘한 얼굴.\n\nLOCATION (lock): A descending maritime-pine trail leading away from the forest slaughterhouse in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The youth killed in the slaughterhouse remains behind with his throat cut; Youngran's claw may retain traces of his blood.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 몸을 반쯤 돌린 채 고산들을 날카롭게 쏘아보는 강영란(흰 머리 여자)의 서늘한 얼굴.\n\nLOCATION (lock): A descending maritime-pine trail leading away from the forest slaughterhouse in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The youth killed in the slaughterhouse remains behind with his throat cut; Youngran's claw may retain traces of his blood.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 몸을 반쯤 돌린 채 고산들을 날카롭게 쏘아보는 강영란(흰 머리 여자)의 서늘한 얼굴.\n\nLOCATION (lock): A descending maritime-pine trail leading away from the forest slaughterhouse in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The youth killed in the slaughterhouse remains behind with his throat cut; Youngran's claw may retain traces of his blood.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S45sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S45sh3.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "허용된 인물 목록에 없는 추가 인물(오른쪽 어깨와 뒷모습)을 화면에 임의로 배치하여 심각한 위반(Hard Violation)에 해당합니다." }, { "label": "B", "score": 7, "verdict_ko": "명시된 단일 인물만 묘사하였으며, 반쯤 몸을 돌린 자세, 피 묻은 손, 낮 시간대의 소나무 숲 배경 등 모든 지침을 충실히 반영했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "허용된 인물 목록에 없는 추가 인물(오른쪽 어깨와 뒷모습)을 화면에 임의로 배치하여 심각한 위반(Hard Violation)에 해당합니다." }, { "label": "B", "score": 7, "verdict_ko": "명시된 단일 인물만 묘사하였으며, 반쯤 몸을 돌린 자세, 피 묻은 손, 낮 시간대의 소나무 숲 배경 등 모든 지침을 충실히 반영했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "몸을 반쯤 돌린 자세와 핏자국이 남은 손 등 지시사항을 정확히 구현했으며, 프레임 내에 허용되지 않은 인물을 추가하지 않아 규칙을 잘 준수했습니다." }, { "label": "B", "score": 3, "verdict_ko": "화면 우측에 지문과 허용 목록에 없는 인물의 뒷모습을 임의로 추가하여 엄격한 인물 등장 제한 규칙을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "몸을 반쯤 돌린 자세와 핏자국이 남은 손 등 지시사항을 정확히 구현했으며, 프레임 내에 허용되지 않은 인물을 추가하지 않아 규칙을 잘 준수했습니다." }, { "label": "A", "score": 3, "verdict_ko": "화면 우측에 지문과 허용 목록에 없는 인물의 뒷모습을 임의로 추가하여 엄격한 인물 등장 제한 규칙을 위반했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "허용된 인물 목록에 없는 추가 인물(오른쪽 어깨와 뒷모습)을 화면에 임의로 배치하여 심각한 위반(Hard Violation)에 해당합니다." }, { "label": "B", "score": 7, "verdict_ko": "명시된 단일 인물만 묘사하였으며, 반쯤 몸을 돌린 자세, 피 묻은 손, 낮 시간대의 소나무 숲 배경 등 모든 지침을 충실히 반영했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S45sh3__bgfirst_bg.png", "bg_asset_id": "366089ca-3869-4446-afc8-94594ed51e1e", "bg_record_key": "S45sh3::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "해송_숲길", "groupbg_asset_id": "dded1895-8314-4902-9d6c-855ef980a51c" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S45sh6": { "input_fingerprint": "5725c00cca7b7ec8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 흙바닥에 두 무릎을 꿇은 채 고개를 푹 숙인 고산들의 무너진 상체.\n\nLOCATION (lock): A secluded dirt section of the forest trail beneath dense pines. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 4 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "두 무릎을 꿇고 고개를 푹 숙인 채 축 늘어진 자세를 완벽히 연출하여 지문의 '무너진 상체'를 가장 훌륭하게 표현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "지문에 명시된 자세를 대체로 따랐으나, 손을 땅에 짚고 지탱하고 있어 상체가 완전히 무너졌다는 묘사에는 다소 미치지 못합니다." }, { "label": "C", "score": 4, "verdict_ko": "고개를 숙이고 무릎을 꿇었으나 양팔을 곧게 펴고 체중을 버티고 있어 '무너진 상체'라는 핵심 지문과 정면으로 배치됩니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S45sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S45sh3" } }, "S45sh8": { "input_fingerprint": "a72541018f1df175", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 바닥에 꿇어앉아 통곡하는 고산들을 두 팔로 따뜻하게 끌어안은 강영란(흰 머리 여자)의 자애로운 뒷모습.\n\nLOCATION (lock): On the forest-trail ground, screened by the surrounding pine woods. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 6, "A": 5 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "완벽한 뒷모습 구도와 감정 표현을 보여주며 남성의 자세도 이전 숏과 잘 이어지나, 여성의 상의 의상이 다르게 렌더링됨." }, { "label": "C", "score": 6, "verdict_ko": "구도와 배경의 연속성은 훌륭하나, 여성에게 원본과 전혀 다른 파란색 코트가 입혀져 의상 오류가 큼." }, { "label": "A", "score": 5, "verdict_ko": "여성에게 남성의 가죽 재킷이 그대로 복제되어 입혀졌으며, 남성이 상체를 세워 안고 있어 이전 숏의 자세 연속성이 떨어짐." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S45sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S45sh6" } }, "S46sh2": { "input_fingerprint": "a3ff565f8c28b983", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 이끼 낀 비석 표면에 '수리영'이라는 글자가 깊게 음각으로 파여 있는 모습의 클로즈업.\n\nLOCATION (lock): The moss-covered face of one of several burial markers scattered through the forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The forest markers continue to bear the carved name “Su Ri-yeong,” with rotting-meat offerings before them. Jihu remains without her necklace.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 8, "B": 5, "C": 6 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "요청된 '수리영' 텍스트가 가장 정확하게 음각으로 구현되었으며, 클로즈업 샷의 의도를 잘 살렸습니다." }, { "label": "B", "score": 5, "verdict_ko": "텍스트의 '리' 자가 심하게 왜곡되어 형태를 알아보기 어렵습니다." }, { "label": "C", "score": 6, "verdict_ko": "배경의 디테일은 좋으나 비석에 새겨진 글자의 형태가 다소 뭉개져 가독성이 떨어집니다." } ], "refs": [], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S46sh12": { "input_fingerprint": "6d04dd498767bd35", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 넘어진 김지후(강수리영) 앞에 우뚝 서서 뻗은 손을 내밀고 있는 고산들의 무표정한 전면.\n\nLOCATION (lock): A forest clearing among the named stone markers, immediately after the animal attack. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same moss-covered grave markers, dense pines, offerings, and forest daylight. Exclude the prior stone-only composition and show the man standing over the fallen woman with an offered hand.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The oversized black goat remains sprawled in the forest, barely breathing after Sandeul struck it; Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "고산들의 무표정한 '전면' 샷 구도를 정확히 연출했으며, 지후의 목걸이가 없는 상태 지시를 완벽히 준수했습니다." }, { "label": "B", "score": 5, "verdict_ko": "지후의 목걸이 제거 상태는 유지했으나, 샷 텍스트에 명시된 고산들의 '전면'이 아닌 측면 각도에서 촬영되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "고산들의 전면 구도를 벗어났으며, 지후가 목걸이를 그대로 착용하고 있어 주요 상태 지시를 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S46sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S46sh2" } }, "S46sh18": { "input_fingerprint": "fbdd8ba51f53b4e2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 기절한 거대한 수리영을 공격한 검은 염소를 어깨 위로 번쩍 둘러멘 고산들의 힘 들어간 넓은 뒷모습.\n\nLOCATION (lock): The descending trail leaving the burial-marker area of the forest. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same pine trunks, mossy stones, rotten offerings, and daylight. Exclude the fallen woman's earlier position and show the man carrying the unconscious black goat away.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The unconscious goat hangs limp across 고산들's shoulder, with its torso supported by the shoulder and its head and four legs dangling down on opposite sides of his body.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The unconscious black goat hangs limp across Sandeul's shoulder, with its torso supported by the shoulder and its head and four legs dangling down on opposite sides of his body, as he leaves the forest. Jihu follows without her bone necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리); 수리영을 공격한 검은 염소 (변이 상태) (거대한 검은 염소, 위협적인 두 뿔, 거친 검은 털, 큼직한 염소 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 3, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "산들의 뒷모습과 염소를 멘 샷 구도를 가장 정확히 구현했으나, 지시문과 달리 쓰러진 여자가 프레임에 남아있어 감점됨." }, { "label": "A", "score": 3, "verdict_ko": "주인공의 머리가 완전히 누락된 물리적으로 불가능한 해부학적 오류(치명적 위반)가 있으며, 텍스트에 없는 인물이 추가됨." }, { "label": "B", "score": 3, "verdict_ko": "A와 마찬가지로 주인공의 머리가 생략된 치명적인 신체 구조 오류가 발생했으며, 배경에 불필요한 인물이 묘사됨." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S46sh12_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 수리영을 공격한 검은 염소 (변이 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S46sh12" } }, "groupbg::육지_항구_선착장": { "input_fingerprint": "8ec817afbe01b7a0", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "육지_항구_선착장", "tags": [ "S100sh3", "S100sh5", "S101sh15", "S101sh6", "S47sh1", "S47sh10", "S47sh12", "S53sh1", "S53sh3" ] }, "context_sig": "495851435ee4649e" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: Beside a barrel fire on a mainland fishing harbor pier under an overcast sky.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n낚시 어선 갑판과 외부: 5톤 규모의 흰색 소형 낚시 어선 외부. 뱃머리와 선미 쪽에 낚시 장비와 사람들이 있다. (특징: 흰색 선체와 갑판; 뱃머리와 선미의 빈 공간; 난간 바(rail bar); 바다가 보이는 배경) / 육지 해안 마을 선착장: 금월도 밖 육지의 항구. 드럼통 모닥불과 어선, 폴리스라인이 쳐진 구역이 있다. (특징: 불타는 드럼통 모닥불; 정박된 배들; 바닥의 훼손된 시체와 노란색 폴리스라인; 바다와 인접한 시멘트 호안)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 해안마을 선착장 - 낮\n- 실외. 해안마을 선착장 - 낮\n- 실외. 해안마을 선착장 - 낮\n- 선착장에 서 있는 창모에게 다가온다.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: Beside a barrel fire on a mainland fishing harbor pier under an overcast sky.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n낚시 어선 갑판과 외부: 5톤 규모의 흰색 소형 낚시 어선 외부. 뱃머리와 선미 쪽에 낚시 장비와 사람들이 있다. (특징: 흰색 선체와 갑판; 뱃머리와 선미의 빈 공간; 난간 바(rail bar); 바다가 보이는 배경) / 육지 해안 마을 선착장: 금월도 밖 육지의 항구. 드럼통 모닥불과 어선, 폴리스라인이 쳐진 구역이 있다. (특징: 불타는 드럼통 모닥불; 정박된 배들; 바닥의 훼손된 시체와 노란색 폴리스라인; 바다와 인접한 시멘트 호안)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 해안마을 선착장 - 낮\n- 실외. 해안마을 선착장 - 낮\n- 실외. 해안마을 선착장 - 낮\n- 선착장에 서 있는 창모에게 다가온다.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_육지_항구_선착장_19d1cf.png", "asset_id": "a146bae0-100d-4928-b618-164368b367c6", "input_asset_ids": [ "a0501868-d04a-4e84-90e9-fcd29b8cba01" ], "origin_tag": "S47sh1", "place_text": "Beside a barrel fire on a mainland fishing harbor pier under an overcast sky.", "origin_inputs": { "place_text": "Beside a barrel fire on a mainland fishing harbor pier under an overcast sky.", "time_of_day_en": "day", "conti_asset_id": "a0501868-d04a-4e84-90e9-fcd29b8cba01" } }, "S47sh1::bgfirst_bg": { "input_fingerprint": "a98cbcb79039a042", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 흐린 하늘 아래 선착장의 드럼통 모닥불 옆에 쪼그려 앉아 수심 가득한 표정으로 불꽃을 응시하는 정인우의 전신.\n\nLOCATION (lock): Beside a barrel fire on a mainland fishing harbor pier under an overcast sky.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a low standing height several meters away, frame 정인우 full-body across the drum-barrel fire as he crouches beside it, placing the flames below his lowered gaze and retaining the cloudy sky above him. Keep the frame static and wide, using the distance between fire and figure to make his worry feel exposed on the pier.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정인우 in the middle-right of the frame, midground, looks toward drum-barrel fire; drum-barrel fire in the lower-center of the frame, foreground; cloudy sky in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: drum barrel fire (Burning beside the seated group) — The open top and flames face upward between camera and 정인우; used as Foreground visual barrier and the object beneath 정인우's eyeline; pier (Occupied by fishermen near the fire) — Its length recedes behind and beside 정인우; used as Open spatial context around the isolated crouching figure; cloudy sky (Cloudy as specified); used as Upper-frame atmosphere retained by the slight upward tilt.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 흐린 하늘 아래 선착장의 드럼통 모닥불 옆에 쪼그려 앉아 수심 가득한 표정으로 불꽃을 응시하는 정인우의 전신.\n\nLOCATION (lock): Beside a barrel fire on a mainland fishing harbor pier under an overcast sky.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a low standing height several meters away, frame 정인우 full-body across the drum-barrel fire as he crouches beside it, placing the flames below his lowered gaze and retaining the cloudy sky above him. Keep the frame static and wide, using the distance between fire and figure to make his worry feel exposed on the pier.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정인우 in the middle-right of the frame, midground, looks toward drum-barrel fire; drum-barrel fire in the lower-center of the frame, foreground; cloudy sky in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: drum barrel fire (Burning beside the seated group) — The open top and flames face upward between camera and 정인우; used as Foreground visual barrier and the object beneath 정인우's eyeline; pier (Occupied by fishermen near the fire) — Its length recedes behind and beside 정인우; used as Open spatial context around the isolated crouching figure; cloudy sky (Cloudy as specified); used as Upper-frame atmosphere retained by the slight upward tilt.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S47sh1__bgfirst_bg.png", "asset_id": "394a24f3-1f7b-48ff-b376-79e87780bab9", "input_asset_ids": [ "a0501868-d04a-4e84-90e9-fcd29b8cba01", "a146bae0-100d-4928-b618-164368b367c6" ] }, "S47sh1": { "input_fingerprint": "b6e71301bf4103a9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 흐린 하늘 아래 선착장의 드럼통 모닥불 옆에 쪼그려 앉아 수심 가득한 표정으로 불꽃을 응시하는 정인우의 전신.\n\nLOCATION (lock): Beside a barrel fire on a mainland fishing harbor pier under an overcast sky. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the cast on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 흐린 하늘 아래 선착장의 드럼통 모닥불 옆에 쪼그려 앉아 수심 가득한 표정으로 불꽃을 응시하는 정인우의 전신.\n\nLOCATION (lock): Beside a barrel fire on a mainland fishing harbor pier under an overcast sky. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the cast on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 흐린 하늘 아래 선착장의 드럼통 모닥불 옆에 쪼그려 앉아 수심 가득한 표정으로 불꽃을 응시하는 정인우의 전신.\n\nLOCATION (lock): Beside a barrel fire on a mainland fishing harbor pier under an overcast sky. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the cast on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S47sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S47sh1.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_육지_항구_선착장_19d1cf.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "프롬프트에 명시되지 않은 다수의 인물들이 배경에 추가되어 심각한 지시 위반(Hard Violation)입니다." }, { "label": "B", "score": 7, "verdict_ko": "지정된 장소에서 명시된 단일 인물(정인우)의 전신, 깁스 착용 상태, 그리고 표정 연기를 정확히 재현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "프롬프트에 명시되지 않은 다수의 인물들이 배경에 추가되어 심각한 지시 위반(Hard Violation)입니다." }, { "label": "B", "score": 7, "verdict_ko": "지정된 장소에서 명시된 단일 인물(정인우)의 전신, 깁스 착용 상태, 그리고 표정 연기를 정확히 재현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 인물 정인우만이 프레임 안에 존재하며, 장소, 자세, 팔 깁스, 복장 모두 지시사항을 훌륭하게 준수했습니다." }, { "label": "B", "score": 3, "verdict_ko": "프롬프트에 명시되지 않은 여러 명의 다른 인물들을 추가로 그려넣어 '절대 다른 사람을 추가하지 말 것'이라는 규칙을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "지정된 인물 정인우만이 프레임 안에 존재하며, 장소, 자세, 팔 깁스, 복장 모두 지시사항을 훌륭하게 준수했습니다." }, { "label": "A", "score": 3, "verdict_ko": "프롬프트에 명시되지 않은 여러 명의 다른 인물들을 추가로 그려넣어 '절대 다른 사람을 추가하지 말 것'이라는 규칙을 위반했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 16 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "프롬프트에 명시되지 않은 다수의 인물들이 배경에 추가되어 심각한 지시 위반(Hard Violation)입니다." }, { "label": "B", "score": 7, "verdict_ko": "지정된 장소에서 명시된 단일 인물(정인우)의 전신, 깁스 착용 상태, 그리고 표정 연기를 정확히 재현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_육지_항구_선착장_19d1cf.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S47sh1__bgfirst_bg.png", "bg_asset_id": "394a24f3-1f7b-48ff-b376-79e87780bab9", "bg_record_key": "S47sh1::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "육지_항구_선착장", "groupbg_asset_id": "a146bae0-100d-4928-b618-164368b367c6" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S47sh10": { "input_fingerprint": "217d9fc1baa9b144", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 시체 옆에서 반쯤 몸을 일으킨 자세로 폴리스라인 밖을 향해 난감한 표정을 짓고 있는 노 형사의 굳은 상체.\n\nLOCATION (lock): Inside a police cordon around a mutilated body on the mainland harbor pier. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The corpse is sprawled on the shore beside the investigators, with the head and one arm completely torn away and the remaining torso and limbs resting motionless on the ground.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The unidentified corpse remains at the shoreline scene with its head and one arm missing and severe animal-like mutilation.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 노 형사 (40대 한국인 남성, 평범한 중년 얼굴, 짧고 단정한 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "캐릭터의 의상(목에 두른 수건 포함)을 레퍼런스와 가장 정확하게 일치시켰으며, 시체 옆에서 반쯤 몸을 일으키는 굳은 상체와 난감한 표정을 훌륭하게 연출했습니다." }, { "label": "B", "score": 5, "verdict_ko": "지정된 자세와 배경은 어느 정도 부합하나, 캐릭터 레퍼런스의 핵심 소품인 목 수건이 누락되고 프롬프트에 없는 붕대가 손에 추가되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "캐릭터의 의상이 지정된 레퍼런스와 전혀 다른 갈색 코트 등으로 잘못 묘사되어 외형 일치도가 크게 떨어집니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S47sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 노 형사: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S47sh1" } }, "S47sh12": { "input_fingerprint": "45141754d595861c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 노 형사를 향해 진지한 눈빛으로 입을 살짝 벌린 채 확신에 찬 표정을 짓고 있는 정인우의 옆얼굴.\n\nLOCATION (lock): At the edge of the police line on the mainland harbor pier. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same police line, gathered onlookers, damaged corpse area, and overcast dock light. Exclude the detective's half-rising pose and frame the injured man addressing him.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The mutilated corpse remains fixed within the police line, missing its head and one arm. Inwoo still has one arm in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 앵글(옆얼굴)과 표정을 잘 담아냈으며, 주인공의 의상(베이지색 코트)과 깁스 상태를 레퍼런스에 맞게 가장 정확히 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "형사를 마주 보는 구도와 배경의 연속성은 좋으나, 주인공 정인우의 의상이 캐릭터 레퍼런스와 전혀 다르게 묘사되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "주인공의 의상이 레퍼런스와 일치하지 않으며, 샷 텍스트에 명시된 대화 상대(노 형사)가 프레임 내에 시각적으로 드러나지 않았습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S47sh10_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S47sh10" } }, "S48sh1::bgfirst_bg": { "input_fingerprint": "e469f379edc68367", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두컴컴한 주방 도마 위, 커다란 식칼을 강하게 내리쳐 검은 염소의 목에 칼날이 깊숙이 박힌 순간의 고산들의 힘 실린 팔.\n\nLOCATION (lock): At the chopping board in the restaurant kitchen, with the hall and open room visible nearby in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Set the camera low beside the cutting board, close to its surface, with the goat's neck held in the lower middle and 고산들's driving forearm descending diagonally from the upper side. Freeze the peak impact as the large knife bites into the neck, retaining enough of his tightened torso and the board edges to preserve realistic scale rather than isolating the blade.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: cutting board (In active use for butchering) — Its top surface is seen at a shallow angle beneath the goat's neck; used as Low ground plane for the impact and scale reference around the neck and knife; kitchen opening (Provides passage toward the dining area) — The opening recedes behind the cutting area toward the dining space; used as Secondary context beyond the impact point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두컴컴한 주방 도마 위, 커다란 식칼을 강하게 내리쳐 검은 염소의 목에 칼날이 깊숙이 박힌 순간의 고산들의 힘 실린 팔.\n\nLOCATION (lock): At the chopping board in the restaurant kitchen, with the hall and open room visible nearby in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Set the camera low beside the cutting board, close to its surface, with the goat's neck held in the lower middle and 고산들's driving forearm descending diagonally from the upper side. Freeze the peak impact as the large knife bites into the neck, retaining enough of his tightened torso and the board edges to preserve realistic scale rather than isolating the blade.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: cutting board (In active use for butchering) — Its top surface is seen at a shallow angle beneath the goat's neck; used as Low ground plane for the impact and scale reference around the neck and knife; kitchen opening (Provides passage toward the dining area) — The opening recedes behind the cutting area toward the dining space; used as Secondary context beyond the impact point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S48sh1__bgfirst_bg.png", "asset_id": "23151f65-a490-4af4-83fb-fefb325b4dd7", "input_asset_ids": [ "b64fa413-b509-4ca6-a451-70b72b32722a", "7708314a-bf99-4888-bd2e-f885cf704b49" ] }, "S48sh1": { "input_fingerprint": "2e7431a84fbdb569", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두컴컴한 주방 도마 위, 커다란 식칼을 강하게 내리쳐 검은 염소의 목에 칼날이 깊숙이 박힌 순간의 고산들의 힘 실린 팔.\n\nLOCATION (lock): At the chopping board in the restaurant kitchen, with the hall and open room visible nearby in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The dead goat is stretched on the kitchen cutting surface with its torso supported horizontally, all four legs slack, and its neck placed beneath the descending cleaver as the head is severed.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The dead goat is stretched on the kitchen cutting surface with its torso supported horizontally, all four legs slack, and its neck placed beneath the descending cleaver as the head is severed, and Sandeul begins butchering its bloody flesh. Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두컴컴한 주방 도마 위, 커다란 식칼을 강하게 내리쳐 검은 염소의 목에 칼날이 깊숙이 박힌 순간의 고산들의 힘 실린 팔.\n\nLOCATION (lock): At the chopping board in the restaurant kitchen, with the hall and open room visible nearby in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The dead goat is stretched on the kitchen cutting surface with its torso supported horizontally, all four legs slack, and its neck placed beneath the descending cleaver as the head is severed.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The dead goat is stretched on the kitchen cutting surface with its torso supported horizontally, all four legs slack, and its neck placed beneath the descending cleaver as the head is severed, and Sandeul begins butchering its bloody flesh. Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두컴컴한 주방 도마 위, 커다란 식칼을 강하게 내리쳐 검은 염소의 목에 칼날이 깊숙이 박힌 순간의 고산들의 힘 실린 팔.\n\nLOCATION (lock): At the chopping board in the restaurant kitchen, with the hall and open room visible nearby in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The dead goat is stretched on the kitchen cutting surface with its torso supported horizontally, all four legs slack, and its neck placed beneath the descending cleaver as the head is severed.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The dead goat is stretched on the kitchen cutting surface with its torso supported horizontally, all four legs slack, and its neck placed beneath the descending cleaver as the head is severed, and Sandeul begins butchering its bloody flesh. Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S48sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S48sh1.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 식칼·도축용 칼·의식용 칼: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B01.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 식칼·도축용 칼·의식용 칼: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "텍스트가 요구한 흑염소 도축 장면 대신 호박을 썰고 있으며, 염소가 살아서 바닥에 서 있는 치명적인 오류가 있습니다." }, { "label": "B", "score": 9, "verdict_ko": "도마 위 흑염소의 목에 칼이 박힌 순간과 고산들의 힘이 실린 팔을 클로즈업하여 텍스트의 구도와 동작을 완벽하게 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "텍스트가 요구한 흑염소 도축 장면 대신 호박을 썰고 있으며, 염소가 살아서 바닥에 서 있는 치명적인 오류가 있습니다." }, { "label": "B", "score": 9, "verdict_ko": "도마 위 흑염소의 목에 칼이 박힌 순간과 고산들의 힘이 실린 팔을 클로즈업하여 텍스트의 구도와 동작을 완벽하게 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지문이 요구한 '힘 실린 팔'과 '염소 목에 박힌 칼'을 정확한 프레이밍과 동작으로 완벽하게 구현했습니다." }, { "label": "B", "score": 2, "verdict_ko": "염소가 바닥에 살아있고 호박을 썰고 있어 텍스트의 핵심인 염소 도축 상황을 전혀 반영하지 못한 치명적 오류가 있습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지문이 요구한 '힘 실린 팔'과 '염소 목에 박힌 칼'을 정확한 프레이밍과 동작으로 완벽하게 구현했습니다." }, { "label": "A", "score": 2, "verdict_ko": "염소가 바닥에 살아있고 호박을 썰고 있어 텍스트의 핵심인 염소 도축 상황을 전혀 반영하지 못한 치명적 오류가 있습니다." } ] }, "combined": { "totals": { "A": 4, "B": 17 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 4, "B": 17 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "텍스트가 요구한 흑염소 도축 장면 대신 호박을 썰고 있으며, 염소가 살아서 바닥에 서 있는 치명적인 오류가 있습니다." }, { "label": "B", "score": 9, "verdict_ko": "도마 위 흑염소의 목에 칼이 박힌 순간과 고산들의 힘이 실린 팔을 클로즈업하여 텍스트의 구도와 동작을 완벽하게 구현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B01.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 식칼·도축용 칼·의식용 칼: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S48sh1__bgfirst_bg.png", "bg_asset_id": "23151f65-a490-4af4-83fb-fefb325b4dd7", "bg_record_key": "S48sh1::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S48sh10": { "input_fingerprint": "8a1709d94ac89911", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 자애로운 미소를 띤 채 뻗은 손을 김지후(강수리영)의 어깨 위에 얹고 정지한 강영란(흰 머리 여자)의 상체.\n\nLOCATION (lock): The table area of the restaurant hall beside the working kitchen. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The butchered black-goat meat and severed head remain in the kitchen while Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 3, "B": 7, "C": 5 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 남성이 배경에 등장하여 추가 인물 금지 규정을 위반했습니다." }, { "label": "B", "score": 7, "verdict_ko": "요구된 상체 구도, 자애로운 표정 및 포즈를 정확히 연출했으며, 레퍼런스의 의상(두건)과 배경 요소까지 가장 충실히 유지했습니다." }, { "label": "C", "score": 5, "verdict_ko": "지시된 구도와 동작은 나타났으나, 강영란의 머리 수건이 누락되었고 식당 배경의 공간 구조가 기준 이미지와 다르게 왜곡되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S48sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S48sh1" } }, "groupbg::해안_횟집_자리": { "input_fingerprint": "5ecee06ced26fecf", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "해안_횟집_자리", "tags": [ "S49sh5", "S49sh7" ] }, "context_sig": "57f21906aa94e519" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: An outdoor table at a coastal seafood restaurant, overlooking the dark night sea.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안 마을 횟집 야외 테이블: 육지 항구 근처, 밤바다가 보이는 식당 야외 데크. (특징: 플라스틱 의자와 테이블; 소주병과 잔; 멀리 보이는 조업 어선들의 불빛)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실내. 해안마을 횟집 - 밤\n- 그 풍경이 보이는 바깥 자리에 앉아 낮에 본 형사 둘과 술을 마시는 인우.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: An outdoor table at a coastal seafood restaurant, overlooking the dark night sea.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안 마을 횟집 야외 테이블: 육지 항구 근처, 밤바다가 보이는 식당 야외 데크. (특징: 플라스틱 의자와 테이블; 소주병과 잔; 멀리 보이는 조업 어선들의 불빛)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실내. 해안마을 횟집 - 밤\n- 그 풍경이 보이는 바깥 자리에 앉아 낮에 본 형사 둘과 술을 마시는 인우.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해안_횟집_자리_81d51c.png", "asset_id": "eb0ff04b-3089-4b55-9015-bea632e27627", "input_asset_ids": [ "9cf04efb-56f9-403b-a902-3c0b1506535e" ], "origin_tag": "S49sh5", "place_text": "An outdoor table at a coastal seafood restaurant, overlooking the dark night sea.", "origin_inputs": { "place_text": "An outdoor table at a coastal seafood restaurant, overlooking the dark night sea.", "time_of_day_en": "night", "conti_asset_id": "9cf04efb-56f9-403b-a902-3c0b1506535e" } }, "S49sh5::bgfirst_bg": { "input_fingerprint": "09e1f60dcb2fc8d7", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): CCTV 캡처 사진이 띄워진 핸드폰 화면을 형사들 앞으로 내민 채 고정된 정인우의 손.\n\nLOCATION (lock): An outdoor table at a coastal seafood restaurant, overlooking the dark night sea.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just above table height, hold a static close view along 정인우’s extended hand, with the phone occupying the lower central third and its bezel preserving the sense of a captured image rather than an unmediated alley. The CCTV frame of 강영란(흰 머리 여자) and 황봉업 remains sharp while 정인우’s wrist and the detectives’ seated forms provide soft scale references at opposite edges; the composition is poised to tilt up along their shared sightline.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: phone displaying the CCTV capture in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 핸드폰 화면 (On, displaying the CCTV capture) — Its front face is angled toward the detectives and camera, visibly carrying the CCTV capture of 강영란 and 황봉업 in a dim alley; used as Primary focal surface held within the lower central third; 테이블 가장자리 (In use during the drinking conversation) — Its near edge runs laterally across the bottom of frame; used as A narrow foreground baseline beneath the extended hand.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): CCTV 캡처 사진이 띄워진 핸드폰 화면을 형사들 앞으로 내민 채 고정된 정인우의 손.\n\nLOCATION (lock): An outdoor table at a coastal seafood restaurant, overlooking the dark night sea.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just above table height, hold a static close view along 정인우’s extended hand, with the phone occupying the lower central third and its bezel preserving the sense of a captured image rather than an unmediated alley. The CCTV frame of 강영란(흰 머리 여자) and 황봉업 remains sharp while 정인우’s wrist and the detectives’ seated forms provide soft scale references at opposite edges; the composition is poised to tilt up along their shared sightline.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: phone displaying the CCTV capture in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 핸드폰 화면 (On, displaying the CCTV capture) — Its front face is angled toward the detectives and camera, visibly carrying the CCTV capture of 강영란 and 황봉업 in a dim alley; used as Primary focal surface held within the lower central third; 테이블 가장자리 (In use during the drinking conversation) — Its near edge runs laterally across the bottom of frame; used as A narrow foreground baseline beneath the extended hand.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S49sh5__bgfirst_bg.png", "asset_id": "e6d9e6f8-985a-45b0-98ed-0cc4d8f4eb88", "input_asset_ids": [ "9cf04efb-56f9-403b-a902-3c0b1506535e", "eb0ff04b-3089-4b55-9015-bea632e27627" ] }, "S49sh5": { "input_fingerprint": "aaba98ee9509af20", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): CCTV 캡처 사진이 띄워진 핸드폰 화면을 형사들 앞으로 내민 채 고정된 정인우의 손.\n\nLOCATION (lock): An outdoor table at a coastal seafood restaurant, overlooking the dark night sea. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to hold up his phone displaying the CCTV capture of Youngran and Hwang outside Minsook's house; his injured arm remains cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): CCTV 캡처 사진이 띄워진 핸드폰 화면을 형사들 앞으로 내민 채 고정된 정인우의 손.\n\nLOCATION (lock): An outdoor table at a coastal seafood restaurant, overlooking the dark night sea. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to hold up his phone displaying the CCTV capture of Youngran and Hwang outside Minsook's house; his injured arm remains cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): CCTV 캡처 사진이 띄워진 핸드폰 화면을 형사들 앞으로 내민 채 고정된 정인우의 손.\n\nLOCATION (lock): An outdoor table at a coastal seafood restaurant, overlooking the dark night sea. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to hold up his phone displaying the CCTV capture of Youngran and Hwang outside Minsook's house; his injured arm remains cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S49sh5__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S49sh5.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해안_횟집_자리_81d51c.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "형사들 앞으로 폰을 내민 정인우의 동작과 깁스한 팔, CCTV 화면 및 배경이 프롬프트 지시대로 안정적으로 구현되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "화면 아래에 해부학적으로 불가능한 위치의 깁스한 팔과 정체를 알 수 없는 손이 얽혀 있어 심각한 신체 왜곡(Hard Violation)이 발생했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "형사들 앞으로 폰을 내민 정인우의 동작과 깁스한 팔, CCTV 화면 및 배경이 프롬프트 지시대로 안정적으로 구현되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "화면 아래에 해부학적으로 불가능한 위치의 깁스한 팔과 정체를 알 수 없는 손이 얽혀 있어 심각한 신체 왜곡(Hard Violation)이 발생했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "폰을 쥔 팔과 깁스를 한 팔의 방향 및 해부학적 연결이 불가능하게 배치되어 심각한 신체 구조 오류(Hard Violation)로 실격입니다." }, { "label": "B", "score": 7, "verdict_ko": "장소, 깁스, 폰 화면 등 모든 요소를 훌륭하게 구현했으나, '정인우의 손'에 집중된 지문보다 샷 크기가 넓게(오버더숄더) 잡혀 일부 감점되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 3, "verdict_ko": "폰을 쥔 팔과 깁스를 한 팔의 방향 및 해부학적 연결이 불가능하게 배치되어 심각한 신체 구조 오류(Hard Violation)로 실격입니다." }, { "label": "A", "score": 7, "verdict_ko": "장소, 깁스, 폰 화면 등 모든 요소를 훌륭하게 구현했으나, '정인우의 손'에 집중된 지문보다 샷 크기가 넓게(오버더숄더) 잡혀 일부 감점되었습니다." } ] }, "combined": { "totals": { "A": 14, "B": 6 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 6 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "형사들 앞으로 폰을 내민 정인우의 동작과 깁스한 팔, CCTV 화면 및 배경이 프롬프트 지시대로 안정적으로 구현되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "화면 아래에 해부학적으로 불가능한 위치의 깁스한 팔과 정체를 알 수 없는 손이 얽혀 있어 심각한 신체 왜곡(Hard Violation)이 발생했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해안_횟집_자리_81d51c.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S49sh5__bgfirst_bg.png", "bg_asset_id": "e6d9e6f8-985a-45b0-98ed-0cc4d8f4eb88", "bg_record_key": "S49sh5::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "해안_횟집_자리", "groupbg_asset_id": "eb0ff04b-3089-4b55-9015-bea632e27627" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S49sh7": { "input_fingerprint": "2c0d620db0c8ddf0", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 놀란 표정으로 서로 시선을 맞춘 채 굳어 있는 신 형사와 조 형사의 옆모습.\n\nLOCATION (lock): The same seafront outdoor dining table under the restaurant’s nighttime light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The CCTV image remains displayed on Inwoo's phone before the two detectives, and his injured arm remains in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 신 형사 (30대 한국인 남성, 또렷한 얼굴, 활동적인 인상, 단정한 짧은 머리); 조 형사 (40대 한국인 남성, 넓은 중년 얼굴, 친근한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "이전 샷의 복장(조 형사의 어두운 셔츠)과 깁스 디테일(검은 끈)을 정확히 유지하며 두 형사가 시선을 맞추는 놀란 옆모습을 훌륭히 연출함." }, { "label": "B", "score": 5, "verdict_ko": "구도와 표정은 적절하나, 조 형사의 셔츠 색상(흰색)과 인우의 깁스 끈 색상 등 연속성 유지 조건에서 감점 요소가 있음." }, { "label": "C", "score": 3, "verdict_ko": "이전 샷의 복장 유지 조건을 무시하고 임의로 변경했으며, 몸체 없이 팔만 허공에 떠 있는 치명적인 물리적 오류가 발생함." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S49sh5_sel.png" }, { "label": "CHARACTER REFERENCE — 신 형사: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 조 형사: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S49sh5" } }, "S50sh2::bgfirst_bg": { "input_fingerprint": "0bf5cb5f6ab7c59b", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 김지후(강수리영)의 시선 너머, 골목길에 우두커니 서서 식당 쪽을 감시하듯 쳐다보는 한국인 마을 청년 둘의 실루엣.\n\nLOCATION (lock): Inside the restaurant’s dark back room at a curtained window overlooking the village lane.\n\nTIME OF DAY (lock): night, near-full moon.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From inside the room at eye level just behind 김지후(강수리영)’s shoulder, begin the pan through the narrow curtain opening and pause on two Korean village youths arranged apart in the alley, both watching toward the restaurant. Her blurred shoulder and lifted curtain edge form the foreground aperture while the youths remain readable as full silhouettes in the middle distance; only their screen position is emphasized relative to the preceding window-side view.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 김지후(강수리영) shoulder in the middle-left of the frame, foreground, looks toward two village youths in the alley; two village youths in the middle-center of the frame, midground, looks toward restaurant window.\n- KEY BACKGROUND ELEMENTS: 커튼 틈 (Slightly opened) — The room-facing folds flank a narrow opening aimed into the alley; used as Foreground aperture that conceals most of the observer; 창문 (Unobstructed within the curtain gap) — The camera sees through it from the dark room toward the two youths in the alley; used as Transparent layer between the hidden observer and the watchers; 골목길 (Occupied by the two watching youths); used as Spatial ground separating the watchers from the restaurant.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 김지후(강수리영)의 시선 너머, 골목길에 우두커니 서서 식당 쪽을 감시하듯 쳐다보는 한국인 마을 청년 둘의 실루엣.\n\nLOCATION (lock): Inside the restaurant’s dark back room at a curtained window overlooking the village lane.\n\nTIME OF DAY (lock): night, near-full moon.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From inside the room at eye level just behind 김지후(강수리영)’s shoulder, begin the pan through the narrow curtain opening and pause on two Korean village youths arranged apart in the alley, both watching toward the restaurant. Her blurred shoulder and lifted curtain edge form the foreground aperture while the youths remain readable as full silhouettes in the middle distance; only their screen position is emphasized relative to the preceding window-side view.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 김지후(강수리영) shoulder in the middle-left of the frame, foreground, looks toward two village youths in the alley; two village youths in the middle-center of the frame, midground, looks toward restaurant window.\n- KEY BACKGROUND ELEMENTS: 커튼 틈 (Slightly opened) — The room-facing folds flank a narrow opening aimed into the alley; used as Foreground aperture that conceals most of the observer; 창문 (Unobstructed within the curtain gap) — The camera sees through it from the dark room toward the two youths in the alley; used as Transparent layer between the hidden observer and the watchers; 골목길 (Occupied by the two watching youths); used as Spatial ground separating the watchers from the restaurant.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S50sh2__bgfirst_bg.png", "asset_id": "ac298d08-9e2a-4cf7-817f-c5d4a3d44c73", "input_asset_ids": [ "62409aca-f712-466f-a019-6932d0f64aab", "57aaaf7f-a278-4a08-8347-c1386e1c793d" ] }, "S50sh2": { "input_fingerprint": "0b8bfce87824b6a5", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, near-full moon.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 시선 너머, 골목길에 우두커니 서서 식당 쪽을 감시하듯 쳐다보는 한국인 마을 청년 둘의 실루엣.\n\nLOCATION (lock): Inside the restaurant’s dark back room at a curtained window overlooking the village lane. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace while the village youths maintain surveillance outside the restaurant.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, near-full moon.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 시선 너머, 골목길에 우두커니 서서 식당 쪽을 감시하듯 쳐다보는 한국인 마을 청년 둘의 실루엣.\n\nLOCATION (lock): Inside the restaurant’s dark back room at a curtained window overlooking the village lane. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace while the village youths maintain surveillance outside the restaurant.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, near-full moon.\n\nSHOT TEXT (authoritative, Korean): 김지후(강수리영)의 시선 너머, 골목길에 우두커니 서서 식당 쪽을 감시하듯 쳐다보는 한국인 마을 청년 둘의 실루엣.\n\nLOCATION (lock): Inside the restaurant’s dark back room at a curtained window overlooking the village lane. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace while the village youths maintain surveillance outside the restaurant.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S50sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S50sh2.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L21B02.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지후의 어깨 너머로 식당을 감시하는 청년들의 시선과 실사 영화의 사실적인 질감을 완벽하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "실사 영화가 아닌 일러스트 스타일로 렌더링되었으며, 골목길의 청년들이 식당 창문이 아닌 다른 곳을 응시하고 있어 지시문과 어긋납니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지후의 어깨 너머로 식당을 감시하는 청년들의 시선과 실사 영화의 사실적인 질감을 완벽하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "실사 영화가 아닌 일러스트 스타일로 렌더링되었으며, 골목길의 청년들이 식당 창문이 아닌 다른 곳을 응시하고 있어 지시문과 어긋납니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 실사 영화 스틸컷 스타일을 잘 충족하며 어깨 너머의 시점과 창밖 청년들의 묘사가 훌륭하지만, 지후에게 목걸이가 묘사된 점은 지시 사항 위반입니다." }, { "label": "B", "score": 3, "verdict_ko": "가장 핵심적인 '실사 사진(photorealistic)' 지시를 무시하고 2D 일러스트 스타일로 렌더링되어 평가 우선순위에서 크게 밀립니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 실사 영화 스틸컷 스타일을 잘 충족하며 어깨 너머의 시점과 창밖 청년들의 묘사가 훌륭하지만, 지후에게 목걸이가 묘사된 점은 지시 사항 위반입니다." }, { "label": "A", "score": 3, "verdict_ko": "가장 핵심적인 '실사 사진(photorealistic)' 지시를 무시하고 2D 일러스트 스타일로 렌더링되어 평가 우선순위에서 크게 밀립니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지후의 어깨 너머로 식당을 감시하는 청년들의 시선과 실사 영화의 사실적인 질감을 완벽하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "실사 영화가 아닌 일러스트 스타일로 렌더링되었으며, 골목길의 청년들이 식당 창문이 아닌 다른 곳을 응시하고 있어 지시문과 어긋납니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L21B02.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S50sh2__bgfirst_bg.png", "bg_asset_id": "ac298d08-9e2a-4cf7-817f-c5d4a3d44c73", "bg_record_key": "S50sh2::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S50sh3": { "input_fingerprint": "3406d0362658fedc", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, near-full moon.\n\nSHOT TEXT (authoritative, Korean): 놀란 얼굴로 커튼 자락을 쥔 채 다급히 반쯤 당겨 닫고 있는 김지후(강수리영)의 긴장된 손과 굳은 표정.\n\nLOCATION (lock): At the window inside the unlit restaurant back room, with moonlight faintly entering around the curtain. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark bedroom, window position, curtain fabric, and youths visible outside under night light. Exclude their static surveillance as the main subject and show the woman urgently closing the curtain.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her necklace as she closes the curtain on the watchers.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 4, "C": 5 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "긴장된 표정으로 커튼을 닫는 지시를 잘 구현했으며 배경 일관성이 뛰어납니다." }, { "label": "B", "score": 4, "verdict_ko": "커튼을 잡은 아래쪽 손의 형태가 해부학적으로 부자연스러워 심각한 오류입니다." }, { "label": "C", "score": 5, "verdict_ko": "이전 샷의 의상은 유지했으나, 창틀 하단 구조가 레퍼런스와 다르게 변형되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S50sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S50sh2" } }, "S51sh2::bgfirst_bg": { "input_fingerprint": "1335690a39a192bb", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 쇠사슬에 묶인 채 초췌한 모습으로 앉아 있는 강민숙(홍은미)을 안타까운 시선으로 내려다보는 고산들의 상체.\n\nLOCATION (lock): A confined chamber deep inside the natural cave, lit by a low-wattage electric bulb.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track at 강민숙(홍은미)’s seated eyeline and angle slightly upward, framing 고산들’s upper body in the center-right while her bound shoulder and chain remain low in the near foreground. His torso bends subtly toward her and his downward eyeline completes the confrontation axis; only his foreground prominence is emphasized over the preceding close view of her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강민숙(홍은미) in the lower-left of the frame, foreground, looks toward 고산들; 고산들 in the middle-right of the frame, midground, looks toward 강민숙(홍은미).\n- KEY BACKGROUND ELEMENTS: 쇠사슬 (Securing 강민숙) — The chain runs from 강민숙’s bound position toward the edge of frame; used as Visible restraint crossing the captive’s foreground body; 촉 낮은 전구 등 (On); used as Established practical source above the confrontation; 동굴 내부 (Occupied by the two characters); used as Compressed enclosure behind 고산들.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 쇠사슬에 묶인 채 초췌한 모습으로 앉아 있는 강민숙(홍은미)을 안타까운 시선으로 내려다보는 고산들의 상체.\n\nLOCATION (lock): A confined chamber deep inside the natural cave, lit by a low-wattage electric bulb.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track at 강민숙(홍은미)’s seated eyeline and angle slightly upward, framing 고산들’s upper body in the center-right while her bound shoulder and chain remain low in the near foreground. His torso bends subtly toward her and his downward eyeline completes the confrontation axis; only his foreground prominence is emphasized over the preceding close view of her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강민숙(홍은미) in the lower-left of the frame, foreground, looks toward 고산들; 고산들 in the middle-right of the frame, midground, looks toward 강민숙(홍은미).\n- KEY BACKGROUND ELEMENTS: 쇠사슬 (Securing 강민숙) — The chain runs from 강민숙’s bound position toward the edge of frame; used as Visible restraint crossing the captive’s foreground body; 촉 낮은 전구 등 (On); used as Established practical source above the confrontation; 동굴 내부 (Occupied by the two characters); used as Compressed enclosure behind 고산들.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S51sh2__bgfirst_bg.png", "asset_id": "96598ff7-5d7a-46b4-bdbf-fab53239b88c", "input_asset_ids": [ "5b39231c-71ab-4e3b-9741-9b95173bcaed", "3d2d81f3-41d9-4b12-98d1-8c3b51938aed" ] }, "S51sh2": { "input_fingerprint": "f4a966de6051693b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 쇠사슬에 묶인 채 초췌한 모습으로 앉아 있는 강민숙(홍은미)을 안타까운 시선으로 내려다보는 고산들의 상체.\n\nLOCATION (lock): A confined chamber deep inside the natural cave, lit by a low-wattage electric bulb. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is seated on the cave floor with her exhausted torso upright and her limbs confined by a heavy chain, facing 고산들 as he stands over her.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains chained in the cave in a weakened, disheveled condition.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 쇠사슬에 묶인 채 초췌한 모습으로 앉아 있는 강민숙(홍은미)을 안타까운 시선으로 내려다보는 고산들의 상체.\n\nLOCATION (lock): A confined chamber deep inside the natural cave, lit by a low-wattage electric bulb. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is seated on the cave floor with her exhausted torso upright and her limbs confined by a heavy chain, facing 고산들 as he stands over her.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains chained in the cave in a weakened, disheveled condition.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 쇠사슬에 묶인 채 초췌한 모습으로 앉아 있는 강민숙(홍은미)을 안타까운 시선으로 내려다보는 고산들의 상체.\n\nLOCATION (lock): A confined chamber deep inside the natural cave, lit by a low-wattage electric bulb. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is seated on the cave floor with her exhausted torso upright and her limbs confined by a heavy chain, facing 고산들 as he stands over her.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains chained in the cave in a weakened, disheveled condition.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S51sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S51sh2.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B03.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지문이 요구한 '고산들의 상체' 프레이밍과 안타깝게 내려다보는 시선을 매우 정확하고 극적으로 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "'상체'라는 숏 크기 지시를 무시하고 인물들의 전신이 거의 다 보이는 풀샷에 가깝게 연출되어 감정선이 약화되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지문이 요구한 '고산들의 상체' 프레이밍과 안타깝게 내려다보는 시선을 매우 정확하고 극적으로 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "'상체'라는 숏 크기 지시를 무시하고 인물들의 전신이 거의 다 보이는 풀샷에 가깝게 연출되어 감정선이 약화되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "여자를 향해 내려다보는 남자의 시선과 자세, 상체 중심의 프레이밍이 지시문과 훌륭하게 일치합니다." }, { "label": "A", "score": 4, "verdict_ko": "남자의 상체가 강조되었으나, 여자를 등지고 엉뚱한 곳을 내려다보고 있어 요구된 상호작용을 충족하지 못했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "여자를 향해 내려다보는 남자의 시선과 자세, 상체 중심의 프레이밍이 지시문과 훌륭하게 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "남자의 상체가 강조되었으나, 여자를 등지고 엉뚱한 곳을 내려다보고 있어 요구된 상호작용을 충족하지 못했습니다." } ] }, "combined": { "totals": { "A": 11, "B": 12 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 12 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지문이 요구한 '고산들의 상체' 프레이밍과 안타깝게 내려다보는 시선을 매우 정확하고 극적으로 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "'상체'라는 숏 크기 지시를 무시하고 인물들의 전신이 거의 다 보이는 풀샷에 가깝게 연출되어 감정선이 약화되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B03.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S51sh2__bgfirst_bg.png", "bg_asset_id": "96598ff7-5d7a-46b4-bdbf-fab53239b88c", "bg_record_key": "S51sh2::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S51sh6": { "input_fingerprint": "8e0542f4b54a8b2d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 멀어지는 고산들의 등을 향해 팽팽해진 쇠사슬을 움켜쥔 채 절규하듯 입을 크게 벌린 강민숙(홍은미)의 처절한 상체.\n\nLOCATION (lock): The chained holding area inside the dim cave, facing the passage leading out. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rough cave walls, low electric light, heavy chains, and damp darkness. Exclude the man's nearby position and show him receding while the chained woman pleads after him.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She remains seated and chained on the cave floor, leaning her upper body toward the departing 고산들 while both hands clutch the chain and pull it taut.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains seated and chained at the same spot on the cave floor as Sandeul leaves, leaning her upper body toward the departing Sandeul while both hands clutch the chain and pull it taut, pleading after him.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "A": 6, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 '처절한 상체' 프레이밍을 가장 정확하게 구현했으며, 사슬을 팽팽하게 쥔 손과 절규하는 표정 등 세부 묘사가 우수함." }, { "label": "A", "score": 6, "verdict_ko": "상체 위주의 구도를 잡았으나 B보다 약간 넓게 프레이밍되었으며 사슬을 쥔 손의 형태가 다소 부자연스러움." }, { "label": "C", "score": 4, "verdict_ko": "'상체'라는 샷 크기 제한을 어기고 전신과 발까지 모두 프레임 안에 포함하여 감점됨." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S51sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S51sh2" } }, "S52sh2::bgfirst_bg": { "input_fingerprint": "8f6c134afc9e8eb6", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 거대한 바위 앞에 두 무릎을 꿇은 채 양손을 뻗어 바위 표면에 바짝 대고 있는 고산들의 굽은 등.\n\nLOCATION (lock): An open ritual clearing deep in the forest, directly before a massive blood-marked rock.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly inward directly behind 고산들 at upper-back height, keeping his bowed back and kneeling legs in the lower center as both arms converge on the giant rock beyond. The rock is held to roughly the upper third rather than overwhelming the human scale, and the blood-marked hand placements lead the eye toward his pressed palms; the single emphasized change from the establishing view is camera distance.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 고산들 in the lower-center of the frame, midground, reaches for giant ritual rock; giant ritual rock in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 거대한 바위 (Marked in many places with dark-red animal blood traces resembling hand smears) — Its broad face is presented toward camera beyond 고산들’s back and hands; used as Ritual focal mass beyond the kneeling figure, limited to the upper portion of frame; 숲속 공터 (Open around the giant rock); used as Spatial boundary around the rock and supplicant; 주변 숲 (Surrounding the clearing at night); used as Vertical enclosure at the clearing’s perimeter.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 거대한 바위 앞에 두 무릎을 꿇은 채 양손을 뻗어 바위 표면에 바짝 대고 있는 고산들의 굽은 등.\n\nLOCATION (lock): An open ritual clearing deep in the forest, directly before a massive blood-marked rock.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly inward directly behind 고산들 at upper-back height, keeping his bowed back and kneeling legs in the lower center as both arms converge on the giant rock beyond. The rock is held to roughly the upper third rather than overwhelming the human scale, and the blood-marked hand placements lead the eye toward his pressed palms; the single emphasized change from the establishing view is camera distance.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 고산들 in the lower-center of the frame, midground, reaches for giant ritual rock; giant ritual rock in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 거대한 바위 (Marked in many places with dark-red animal blood traces resembling hand smears) — Its broad face is presented toward camera beyond 고산들’s back and hands; used as Ritual focal mass beyond the kneeling figure, limited to the upper portion of frame; 숲속 공터 (Open around the giant rock); used as Spatial boundary around the rock and supplicant; 주변 숲 (Surrounding the clearing at night); used as Vertical enclosure at the clearing’s perimeter.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S52sh2__bgfirst_bg.png", "asset_id": "ce498cba-6c24-4576-a568-69ff5edee858", "input_asset_ids": [ "393db544-8fc5-4474-a5f8-f9c8837077bb", "68c88d61-f5d8-4e78-8c60-d1e42e155653" ] }, "S52sh2": { "input_fingerprint": "80a9997ac7bf1dcf", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 거대한 바위 앞에 두 무릎을 꿇은 채 양손을 뻗어 바위 표면에 바짝 대고 있는 고산들의 굽은 등.\n\nLOCATION (lock): An open ritual clearing deep in the forest, directly before a massive blood-marked rock. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The giant ritual rock retains the dark-red animal-blood smears across its surface while Sandeul kneels before it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 거대한 바위 앞에 두 무릎을 꿇은 채 양손을 뻗어 바위 표면에 바짝 대고 있는 고산들의 굽은 등.\n\nLOCATION (lock): An open ritual clearing deep in the forest, directly before a massive blood-marked rock. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The giant ritual rock retains the dark-red animal-blood smears across its surface while Sandeul kneels before it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 거대한 바위 앞에 두 무릎을 꿇은 채 양손을 뻗어 바위 표면에 바짝 대고 있는 고산들의 굽은 등.\n\nLOCATION (lock): An open ritual clearing deep in the forest, directly before a massive blood-marked rock. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The giant ritual rock retains the dark-red animal-blood smears across its surface while Sandeul kneels before it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S52sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S52sh2.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION STRUCTURE PHOTOGRAPH — the confirmed photograph of this exact place and its fixed structure: it is the SINGLE authority for the location, the structure's shape, proportions, materials, colors, openings and every permanent site detail. Never copy its camera framing, time of day or lighting — the shot text is the authority for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_forest_ritual_clearing_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "인물이 바위에서 멀리 떨어져 허공에 손을 들고 있어 '바위 표면에 바짝 대고 있는' 핵심 동작 지시를 위반했습니다." }, { "label": "B", "score": 7, "verdict_ko": "어두운 숲속 피 묻은 바위 앞에 꿇어앉아 바위 표면에 양손을 밀착한 인물의 굽은 등을 지시문과 정확하게 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "인물이 바위에서 멀리 떨어져 허공에 손을 들고 있어 '바위 표면에 바짝 대고 있는' 핵심 동작 지시를 위반했습니다." }, { "label": "B", "score": 7, "verdict_ko": "어두운 숲속 피 묻은 바위 앞에 꿇어앉아 바위 표면에 양손을 밀착한 인물의 굽은 등을 지시문과 정확하게 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "바위 표면에 양손을 바짝 대고 무릎을 꿇은 행동 지문을 정확히 연출했으며, 장소의 디테일과 핏자국, 캐릭터의 착장 모두 기준에 잘 부합합니다." }, { "label": "B", "score": 4, "verdict_ko": "인물이 바위와 멀리 떨어져 있어 '양손을 바위 표면에 바짝 대고 있는' 핵심 행동 지문을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "바위 표면에 양손을 바짝 대고 무릎을 꿇은 행동 지문을 정확히 연출했으며, 장소의 디테일과 핏자국, 캐릭터의 착장 모두 기준에 잘 부합합니다." }, { "label": "A", "score": 4, "verdict_ko": "인물이 바위와 멀리 떨어져 있어 '양손을 바위 표면에 바짝 대고 있는' 핵심 행동 지문을 위반했습니다." } ] }, "combined": { "totals": { "A": 8, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 8, "B": 15 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "인물이 바위에서 멀리 떨어져 허공에 손을 들고 있어 '바위 표면에 바짝 대고 있는' 핵심 동작 지시를 위반했습니다." }, { "label": "B", "score": 7, "verdict_ko": "어두운 숲속 피 묻은 바위 앞에 꿇어앉아 바위 표면에 양손을 밀착한 인물의 굽은 등을 지시문과 정확하게 구현했습니다." } ], "refs": [ { "label": "LOCATION STRUCTURE PHOTOGRAPH — the confirmed photograph of this exact place and its fixed structure: it is the SINGLE authority for the location, the structure's shape, proportions, materials, colors, openings and every permanent site detail. Never copy its camera framing, time of day or lighting — the shot text is the authority for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_forest_ritual_clearing_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S52sh2__bgfirst_bg.png", "bg_asset_id": "ce498cba-6c24-4576-a568-69ff5edee858", "bg_record_key": "S52sh2::bgfirst_bg", "chain_winner": false, "authority": "seed_bg" }, "ref_mode": "seed-bg+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S53sh1::bgfirst_bg": { "input_fingerprint": "a6abbb75054c7936", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 낚시꾼 복장으로 선착장에 서서 어선 위를 잔뜩 찌푸린 눈으로 올려다보는 정인우의 전신.\n\nLOCATION (lock): The boarding edge of a mainland fishing pier beside a moored boat.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height on the dock, hold a static wide three-quarter view of 정인우’s full fishing-costumed figure below the boat deck, his head tilted up toward 황봉업. Keep 황봉업 smaller on the raised deck so the vertical confrontation reads immediately, with dock and vessel establishing scale; the stage-entry composition remains deliberately still.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정인우 in the lower-left of the frame, midground, looks toward 황봉업 on the raised boat deck; 황봉업 in the upper-right of the frame, background, looks toward 정인우 on the dock; raised boat deck in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 선착장 (Occupied by 정인우) — Its working edge runs toward the boat and beneath 정인우; used as Lower plane supporting 정인우’s full figure; 낚시 어선 갑판 (Occupied by 황봉업) — The side and upper deck face the camera above 정인우’s eyeline; used as Raised opposing plane that establishes 황봉업’s height advantage; 어선 조타실 (Part of the moored confrontation setting) — Its dock-facing side is visible behind the raised deck; used as Upper structural mass behind 황봉업.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 낚시꾼 복장으로 선착장에 서서 어선 위를 잔뜩 찌푸린 눈으로 올려다보는 정인우의 전신.\n\nLOCATION (lock): The boarding edge of a mainland fishing pier beside a moored boat.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height on the dock, hold a static wide three-quarter view of 정인우’s full fishing-costumed figure below the boat deck, his head tilted up toward 황봉업. Keep 황봉업 smaller on the raised deck so the vertical confrontation reads immediately, with dock and vessel establishing scale; the stage-entry composition remains deliberately still.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 정인우 in the lower-left of the frame, midground, looks toward 황봉업 on the raised boat deck; 황봉업 in the upper-right of the frame, background, looks toward 정인우 on the dock; raised boat deck in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 선착장 (Occupied by 정인우) — Its working edge runs toward the boat and beneath 정인우; used as Lower plane supporting 정인우’s full figure; 낚시 어선 갑판 (Occupied by 황봉업) — The side and upper deck face the camera above 정인우’s eyeline; used as Raised opposing plane that establishes 황봉업’s height advantage; 어선 조타실 (Part of the moored confrontation setting) — Its dock-facing side is visible behind the raised deck; used as Upper structural mass behind 황봉업.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S53sh1__bgfirst_bg.png", "asset_id": "ee5e2c3d-9801-4f0b-ab39-5851900c3d2d", "input_asset_ids": [ "ea6da879-773b-4cff-8ac9-492206d631ad", "ef8e2161-3434-4687-b5f7-c4f2dc082bb6" ] }, "S53sh1": { "input_fingerprint": "fc815c8baf515b74", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낚시꾼 복장으로 선착장에 서서 어선 위를 잔뜩 찌푸린 눈으로 올려다보는 정인우의 전신.\n\nLOCATION (lock): The boarding edge of a mainland fishing pier beside a moored boat. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear a cast on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낚시꾼 복장으로 선착장에 서서 어선 위를 잔뜩 찌푸린 눈으로 올려다보는 정인우의 전신.\n\nLOCATION (lock): The boarding edge of a mainland fishing pier beside a moored boat. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear a cast on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낚시꾼 복장으로 선착장에 서서 어선 위를 잔뜩 찌푸린 눈으로 올려다보는 정인우의 전신.\n\nLOCATION (lock): The boarding edge of a mainland fishing pier beside a moored boat. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear a cast on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S53sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S53sh1.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION STRUCTURE PHOTOGRAPH — the confirmed photograph of this exact place and its fixed structure: it is the SINGLE authority for the location, the structure's shape, proportions, materials, colors, openings and every permanent site detail. Never copy its camera framing, time of day or lighting — the shot text is the authority for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_mainland_harbor_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "지정된 인물, 깁스 설정, 장소 디테일(표지판 등)을 훌륭하게 구현했으나, 어선 위를 올려다보는 시선 방향이 어긋났습니다." }, { "label": "A", "score": 3, "verdict_ko": "어선을 올려다보는 연출은 시도되었으나, 지문에서 언급되지 않은 정체불명의 인물이 어선 위에 추가된 치명적인 규정 위반이 있습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "지정된 인물, 깁스 설정, 장소 디테일(표지판 등)을 훌륭하게 구현했으나, 어선 위를 올려다보는 시선 방향이 어긋났습니다." }, { "label": "A", "score": 3, "verdict_ko": "어선을 올려다보는 연출은 시도되었으나, 지문에서 언급되지 않은 정체불명의 인물이 어선 위에 추가된 치명적인 규정 위반이 있습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지정된 로케이션과 캐릭터의 외형 및 깁스 설정은 매우 훌륭하게 구현했으나, 어선 위를 올려다보는 시선과 방향이 지문과 어긋납니다." }, { "label": "B", "score": 3, "verdict_ko": "지문에 없는 인물이 어선 위에 등장한 하드 비올레이션이 있으며, 고정된 선착장의 배경 구조물 또한 전혀 일치하지 않습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "지정된 로케이션과 캐릭터의 외형 및 깁스 설정은 매우 훌륭하게 구현했으나, 어선 위를 올려다보는 시선과 방향이 지문과 어긋납니다." }, { "label": "A", "score": 3, "verdict_ko": "지문에 없는 인물이 어선 위에 등장한 하드 비올레이션이 있으며, 고정된 선착장의 배경 구조물 또한 전혀 일치하지 않습니다." } ] }, "combined": { "totals": { "A": 6, "B": 12 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 12 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "지정된 인물, 깁스 설정, 장소 디테일(표지판 등)을 훌륭하게 구현했으나, 어선 위를 올려다보는 시선 방향이 어긋났습니다." }, { "label": "A", "score": 3, "verdict_ko": "어선을 올려다보는 연출은 시도되었으나, 지문에서 언급되지 않은 정체불명의 인물이 어선 위에 추가된 치명적인 규정 위반이 있습니다." } ], "refs": [ { "label": "LOCATION STRUCTURE PHOTOGRAPH — the confirmed photograph of this exact place and its fixed structure: it is the SINGLE authority for the location, the structure's shape, proportions, materials, colors, openings and every permanent site detail. Never copy its camera framing, time of day or lighting — the shot text is the authority for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_mainland_harbor_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S53sh1__bgfirst_bg.png", "bg_asset_id": "ee5e2c3d-9801-4f0b-ab39-5851900c3d2d", "bg_record_key": "S53sh1::bgfirst_bg", "chain_winner": false, "authority": "seed_bg" }, "ref_mode": "seed-bg+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S53sh3": { "input_fingerprint": "df4e424ae374d870", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어금니를 꽉 깨문 채 황봉업(황 씨·민숙 집의 거구 남자)을 향해 경찰 신분증을 불쑥 치켜든 정인우의 단호한 팔.\n\nLOCATION (lock): On the mainland pier directly alongside the fishing boat’s rail. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dock surface, moored fishing boat, gray daylight, and boat operator's position. Exclude the passive negotiation pose and show the man thrusting up his police identification.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo keeps his injured arm in its cast while presenting his police identification.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "어금니를 깨문 표정과 다친 왼팔의 연속성을 유지하며 신분증을 상대방을 향해 올바르게 제시했습니다." }, { "label": "B", "score": 5, "verdict_ko": "부상 위치는 일치하나, 신분증의 앞면이 상대방이 아닌 카메라를 향하고 있어 소품 방향 지침을 위반했습니다." }, { "label": "A", "score": 3, "verdict_ko": "이전 장면에서 부상당했던 왼팔이 오른팔로 뒤바뀌어 나타나는 치명적인 신체 연속성 오류(Hard Violation)가 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S53sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S53sh1" }, "lane_policy": "ab_select_bypass:prev" }, "S54sh3::bgfirst_bg": { "input_fingerprint": "2b869fde905ab8e9", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 조타실 안에서 둥근 핸들을 잡은 채 뱃머리에 선 정인우를 죽일 듯이 노려보는 황봉업(황 씨·민숙 집의 거구 남자)의 살벌한 얼굴.\n\nLOCATION (lock): Inside the fishing boat’s wheelhouse, looking through its windows toward the exposed bow in daylight.\n\nTIME OF DAY (lock): day, foggy.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the wheelhouse at 황봉업’s eye level, finish the dolly-in on his murderous three-quarter glare while preserving both hands clamped around the round wheel in the lower third. Through the forward opening, 정인우 remains a small back-facing figure on the bow along 황봉업’s sightline, with fog obscuring the distance; the emphasized change is the tightening camera distance onto 황봉업.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 황봉업 in the middle-left of the frame, foreground, looks toward 정인우 on the bow; 정인우 in the middle-right of the frame, background; boat bow in the lower-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 둥근 조타 핸들 (Held firmly by 황봉업) — Its wheel face is angled toward camera beside the steering position; used as Lower-frame circular anchor containing both of 황봉업’s gripping hands; 조타실 (Occupied by 황봉업) — The steering side faces camera while the forward opening remains behind him; used as Interior enclosure framing 황봉업’s face and forward glare; 어선 뱃머리 (Moving forward through the sea) — It extends forward from the wheelhouse with 정인우 positioned at its end; used as Long depth anchor holding 정인우 beyond the wheelhouse; 바다 안개 (Dense across the sea); used as Obscures the forward view beyond 정인우.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 조타실 안에서 둥근 핸들을 잡은 채 뱃머리에 선 정인우를 죽일 듯이 노려보는 황봉업(황 씨·민숙 집의 거구 남자)의 살벌한 얼굴.\n\nLOCATION (lock): Inside the fishing boat’s wheelhouse, looking through its windows toward the exposed bow in daylight.\n\nTIME OF DAY (lock): day, foggy.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the wheelhouse at 황봉업’s eye level, finish the dolly-in on his murderous three-quarter glare while preserving both hands clamped around the round wheel in the lower third. Through the forward opening, 정인우 remains a small back-facing figure on the bow along 황봉업’s sightline, with fog obscuring the distance; the emphasized change is the tightening camera distance onto 황봉업.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 황봉업 in the middle-left of the frame, foreground, looks toward 정인우 on the bow; 정인우 in the middle-right of the frame, background; boat bow in the lower-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 둥근 조타 핸들 (Held firmly by 황봉업) — Its wheel face is angled toward camera beside the steering position; used as Lower-frame circular anchor containing both of 황봉업’s gripping hands; 조타실 (Occupied by 황봉업) — The steering side faces camera while the forward opening remains behind him; used as Interior enclosure framing 황봉업’s face and forward glare; 어선 뱃머리 (Moving forward through the sea) — It extends forward from the wheelhouse with 정인우 positioned at its end; used as Long depth anchor holding 정인우 beyond the wheelhouse; 바다 안개 (Dense across the sea); used as Obscures the forward view beyond 정인우.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S54sh3__bgfirst_bg.png", "asset_id": "6fa91a8c-ecb1-4d55-abb6-59185a9542cb", "input_asset_ids": [ "d368d438-7739-4191-83fc-bc7b6ce6e092", "34f46a9b-0204-4db3-9122-dc15e14dc9c9", "e6f2c697-eafa-4ffc-b339-d0a0588aecb4" ] }, "S54sh3": { "input_fingerprint": "6b4d9d5bb91e427a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, foggy.\n\nSHOT TEXT (authoritative, Korean): 조타실 안에서 둥근 핸들을 잡은 채 뱃머리에 선 정인우를 죽일 듯이 노려보는 황봉업(황 씨·민숙 집의 거구 남자)의 살벌한 얼굴.\n\nLOCATION (lock): Inside the fishing boat’s wheelhouse, looking through its windows toward the exposed bow in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, foggy.\n\nSHOT TEXT (authoritative, Korean): 조타실 안에서 둥근 핸들을 잡은 채 뱃머리에 선 정인우를 죽일 듯이 노려보는 황봉업(황 씨·민숙 집의 거구 남자)의 살벌한 얼굴.\n\nLOCATION (lock): Inside the fishing boat’s wheelhouse, looking through its windows toward the exposed bow in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, foggy.\n\nSHOT TEXT (authoritative, Korean): 조타실 안에서 둥근 핸들을 잡은 채 뱃머리에 선 정인우를 죽일 듯이 노려보는 황봉업(황 씨·민숙 집의 거구 남자)의 살벌한 얼굴.\n\nLOCATION (lock): Inside the fishing boat’s wheelhouse, looking through its windows toward the exposed bow in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S54sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S54sh3.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L16B03.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_fishing_boat_sel.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "조타실 내부에서 뱃머리를 향하는 앵글, 핸들을 잡고 노려보는 황봉업의 표정과 인상착의가 지시문 및 레퍼런스와 훌륭하게 일치합니다." }, { "label": "B", "score": 3, "verdict_ko": "카메라가 조타실 외부에 위치하여 로케이션 지시를 위반했으며, 갑판에 존재하지 않는 조타 기구가 추가되어 공간 구조를 심각하게 훼손했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "조타실 내부에서 뱃머리를 향하는 앵글, 핸들을 잡고 노려보는 황봉업의 표정과 인상착의가 지시문 및 레퍼런스와 훌륭하게 일치합니다." }, { "label": "B", "score": 3, "verdict_ko": "카메라가 조타실 외부에 위치하여 로케이션 지시를 위반했으며, 갑판에 존재하지 않는 조타 기구가 추가되어 공간 구조를 심각하게 훼손했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "카메라가 조타실 외부에 위치하여 '조타실 내부에서 뱃머리를 바라보는' 공간 지침을 심각하게 위반했습니다." }, { "label": "B", "score": 7, "verdict_ko": "조타실 내부에서 뱃머리를 향하는 올바른 카메라 구도와, 핸들을 잡고 노려보는 인물의 표정 및 위치 관계를 충실히 구현했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 3, "verdict_ko": "카메라가 조타실 외부에 위치하여 '조타실 내부에서 뱃머리를 바라보는' 공간 지침을 심각하게 위반했습니다." }, { "label": "A", "score": 7, "verdict_ko": "조타실 내부에서 뱃머리를 향하는 올바른 카메라 구도와, 핸들을 잡고 노려보는 인물의 표정 및 위치 관계를 충실히 구현했습니다." } ] }, "combined": { "totals": { "A": 15, "B": 6 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 15, "B": 6 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "조타실 내부에서 뱃머리를 향하는 앵글, 핸들을 잡고 노려보는 황봉업의 표정과 인상착의가 지시문 및 레퍼런스와 훌륭하게 일치합니다." }, { "label": "B", "score": 3, "verdict_ko": "카메라가 조타실 외부에 위치하여 로케이션 지시를 위반했으며, 갑판에 존재하지 않는 조타 기구가 추가되어 공간 구조를 심각하게 훼손했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L16B03.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_fishing_boat_sel.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S54sh3__bgfirst_bg.png", "bg_asset_id": "6fa91a8c-ecb1-4d55-abb6-59185a9542cb", "bg_record_key": "S54sh3::bgfirst_bg", "chain_winner": true, "authority": "plate", "seed_attached": true }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S54sh4": { "input_fingerprint": "55bf368c04404188", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, foggy.\n\nSHOT TEXT (authoritative, Korean): 짙은 안개 틈 사이로 어둡고 스산한 기운을 내뿜는 금월도의 기괴한 실루엣이 일부 드러난 정경.\n\nLOCATION (lock): Fogbound open water immediately off the isolated island’s coast. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "C": 6, "A": 5 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "짙은 안개 틈 사이로 섬의 기괴한 실루엣이 일부만 드러나는 프롬프트의 묘사를 가장 극적이고 충실하게 구현했습니다." }, { "label": "C", "score": 6, "verdict_ko": "안개로 덮인 섬의 모습은 보여주나, '안개 틈 사이로'라기보다는 일직선으로 깔린 안개층에 가까워 자연스러움이 떨어집니다." }, { "label": "A", "score": 5, "verdict_ko": "섬 전체의 형태가 너무 선명하게 보여 '일부 드러난' 실루엣이라는 프롬프트의 핵심적인 연출 의도와 거리가 있습니다." } ], "refs": [], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S55sh3::bgfirst_bg": { "input_fingerprint": "bf0f1d57f6d0975c", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 선착장에 모여 적대적인 눈빛으로 정인우를 노려보는 한국인 마을 사람들의 굳은 얼굴.\n\nLOCATION (lock): The island pier’s landing area between the moored boat and the village approach.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At head height a few meters in front of the gathering, let the continuing pan settle slightly closer on a compressed band of rigid Korean villagers, while 정인우 remains just behind the camera axis and outside the image. Faces overlap across the middle of the frame at varied three-quarter and profile angles, creating a collective wall of hostile attention rather than a formal lineup.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 금월도 선착장 (Occupied by villagers awaiting 정인우); used as Provides shallow spatial context behind the concentrated row of faces; 마을 쪽 진입로 (정인우 has entered from the dock side); used as Background route aligned with the villagers’ hostile eyelines.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 선착장에 모여 적대적인 눈빛으로 정인우를 노려보는 한국인 마을 사람들의 굳은 얼굴.\n\nLOCATION (lock): The island pier’s landing area between the moored boat and the village approach.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At head height a few meters in front of the gathering, let the continuing pan settle slightly closer on a compressed band of rigid Korean villagers, while 정인우 remains just behind the camera axis and outside the image. Faces overlap across the middle of the frame at varied three-quarter and profile angles, creating a collective wall of hostile attention rather than a formal lineup.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 금월도 선착장 (Occupied by villagers awaiting 정인우); used as Provides shallow spatial context behind the concentrated row of faces; 마을 쪽 진입로 (정인우 has entered from the dock side); used as Background route aligned with the villagers’ hostile eyelines.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S55sh3__bgfirst_bg.png", "asset_id": "5452620c-709f-45bd-a48b-056fef204a57", "input_asset_ids": [ "b5814425-d19e-4f4c-b900-264aabaec18b", "338f96b2-4333-4b47-a7a4-51ae2225f29e" ] }, "S55sh3": { "input_fingerprint": "64db9f1af27e2b50", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장에 모여 적대적인 눈빛으로 정인우를 노려보는 한국인 마을 사람들의 굳은 얼굴.\n\nLOCATION (lock): The island pier’s landing area between the moored boat and the village approach. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the cast on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장에 모여 적대적인 눈빛으로 정인우를 노려보는 한국인 마을 사람들의 굳은 얼굴.\n\nLOCATION (lock): The island pier’s landing area between the moored boat and the village approach. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the cast on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장에 모여 적대적인 눈빛으로 정인우를 노려보는 한국인 마을 사람들의 굳은 얼굴.\n\nLOCATION (lock): The island pier’s landing area between the moored boat and the village approach. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear the cast on his injured arm.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S55sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S55sh3.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "프롬프트 텍스트('강수리영')가 선박에 유출된 것은 치명적 오류이나, 카메라를 노려보는 마을 사람들의 굳은 표정과 시선 처리는 지시문의 의도를 정확히 구현함." }, { "label": "A", "score": 3, "verdict_ko": "깁스를 한 인물이 중복으로 생성되었고, 정인우의 얼굴이 군중 속에 배치되는 등 인물 구성과 시선 처리가 지시문의 상황을 완전히 왜곡함." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "프롬프트 텍스트('강수리영')가 선박에 유출된 것은 치명적 오류이나, 카메라를 노려보는 마을 사람들의 굳은 표정과 시선 처리는 지시문의 의도를 정확히 구현함." }, { "label": "A", "score": 3, "verdict_ko": "깁스를 한 인물이 중복으로 생성되었고, 정인우의 얼굴이 군중 속에 배치되는 등 인물 구성과 시선 처리가 지시문의 상황을 완전히 왜곡함." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "정인우와 강영란의 외모를 훌륭히 재현하고 적대적인 분위기를 잘 살렸으나, 엄격한 인물 제한 규정을 어기고 다수의 엑스트라를 포함했으며 깁스 설정이 타인에게 적용되어 아쉽습니다." }, { "label": "A", "score": 3, "verdict_ko": "프롬프트의 인물 이름('강수리영')이 선박 표면에 그대로 인쇄되어 나타나는 치명적인 텍스트 유출 위반이 발생하여 실격입니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "정인우와 강영란의 외모를 훌륭히 재현하고 적대적인 분위기를 잘 살렸으나, 엄격한 인물 제한 규정을 어기고 다수의 엑스트라를 포함했으며 깁스 설정이 타인에게 적용되어 아쉽습니다." }, { "label": "B", "score": 3, "verdict_ko": "프롬프트의 인물 이름('강수리영')이 선박 표면에 그대로 인쇄되어 나타나는 치명적인 텍스트 유출 위반이 발생하여 실격입니다." } ] }, "combined": { "totals": { "A": 9, "B": 7 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 9, "B": 7 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "프롬프트 텍스트('강수리영')가 선박에 유출된 것은 치명적 오류이나, 카메라를 노려보는 마을 사람들의 굳은 표정과 시선 처리는 지시문의 의도를 정확히 구현함." }, { "label": "A", "score": 3, "verdict_ko": "깁스를 한 인물이 중복으로 생성되었고, 정인우의 얼굴이 군중 속에 배치되는 등 인물 구성과 시선 처리가 지시문의 상황을 완전히 왜곡함." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S55sh3__bgfirst_bg.png", "bg_asset_id": "5452620c-709f-45bd-a48b-056fef204a57", "bg_record_key": "S55sh3::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S55sh4": { "input_fingerprint": "466c760c5a2362ee", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 신호가 잡히지 않는 핸드폰 화면을 내려다보며 미간을 찌푸린 정인우의 당황한 얼굴.\n\nLOCATION (lock): The open island pier just after disembarkation in daylight. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dock paving, harbor edge, village backdrop, and daylight. Exclude the welcoming crowd and focus on the man checking his dead phone signal.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo's injured arm remains in a cast as he checks the dead phone signal.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 2 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "'얼굴'을 강조한 클로즈업 프레이밍이 지시문에 가장 부합하며, 이전 샷의 배경과 의상을 충실히 유지했습니다." }, { "label": "A", "score": 5, "verdict_ko": "요구된 얼굴 중심의 샷보다 프레임이 넓게 잡혔으며, 핸드폰 화면에 형태가 뭉개진 임의의 텍스트가 생성되었습니다." }, { "label": "B", "score": 2, "verdict_ko": "이전 샷에 존재하지 않던 선박과 텍스트(성광호)를 배경에 임의로 생성하여 심각한 지시문 위반입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S55sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S55sh3" } }, "S55sh6": { "input_fingerprint": "f3247d6a9b2e339f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 식당 앞에서 강영란(흰 머리 여자) 뒤에 선 채 정인우를 향해 한 손을 들어 반갑게 흔드는 김지후(강수리영)의 상체.\n\nLOCATION (lock): The village-side end of the pier, with the small restaurant entrance behind it. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dock-to-village sightline, colorful houses, harbor light, and daytime look. Exclude the earlier arrival crowd and show the woman waving from behind the elder near the restaurant.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her arrowhead-shaped bone necklace as she greets Inwoo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 프레이밍(상체)과 인물 배치(강영란 뒤)를 정확히 구현했으며, 뼈 목걸이 제외 조건과 이전 샷의 배경 연속성도 가장 잘 지켰습니다." }, { "label": "B", "score": 5, "verdict_ko": "상체 프레이밍과 위치 배치는 적절하나, 뒷배경의 건물이 이전 샷과 전혀 다른 형태와 색상(파란색 차양)으로 바뀌어 장소 일관성을 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "지시된 상체보다 프레이밍이 넓고 두 인물이 나란히 서 있으며, 명시적으로 제외해야 할 화살촉 목걸이를 착용하고 있어 주요 조건들을 어겼습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S55sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S55sh4" }, "lane_policy": "ab_select_bypass:prev" }, "groupbg::섬_해안_바위길": { "input_fingerprint": "e05f3c743ec04d80", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "섬_해안_바위길", "tags": [ "S56sh3", "S56sh5", "S60sh4", "S60sh5" ] }, "context_sig": "b54b6c4004efd7b3" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A narrow rocky coastal path between the island village and the forest.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 해안 바윗길: 바다와 맞닿아 있는 가파르고 거친 암석 지대의 비좁은 길. (특징: 울퉁불퉁한 갯바위 덩어리들; 발아래 출렁이는 바닷물; 가파른 경사)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 해안길 - 낮\n- 길이 사라지며 가파른 바위들이 이어진다.\n- 실외. 해안길 - 낮\n- 다시 해안가 바위길로 나온 수리영과 인우\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: A narrow rocky coastal path between the island village and the forest.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 해안 바윗길: 바다와 맞닿아 있는 가파르고 거친 암석 지대의 비좁은 길. (특징: 울퉁불퉁한 갯바위 덩어리들; 발아래 출렁이는 바닷물; 가파른 경사)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 해안길 - 낮\n- 길이 사라지며 가파른 바위들이 이어진다.\n- 실외. 해안길 - 낮\n- 다시 해안가 바위길로 나온 수리영과 인우\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_섬_해안_바위길_446206.png", "asset_id": "f012a978-a0b0-49d3-8522-e8e0e14139ef", "input_asset_ids": [ "f69df416-f47b-485d-8470-a711c431ca17" ], "origin_tag": "S56sh3", "place_text": "A narrow rocky coastal path between the island village and the forest.", "origin_inputs": { "place_text": "A narrow rocky coastal path between the island village and the forest.", "time_of_day_en": "day", "conti_asset_id": "f69df416-f47b-485d-8470-a711c431ca17" } }, "S56sh3::bgfirst_bg": { "input_fingerprint": "5c38cafc35f58871", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 정인우의 의심 어린 시선에 입술을 굳게 다물고 싸늘하게 식어버린 김지후(강수리영)의 굳은 표정.\n\nLOCATION (lock): A narrow rocky coastal path between the island village and the forest.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Finish the eye-level dolly-in on 김지후(강수리영)’s front three-quarter side, isolating her firmly closed lips and suddenly frozen expression while 정인우 remains beyond the frame on her eyeline. Her face occupies most of the image, with only a narrow trace of the coastal path retained at the edge before the camera is ready to ease past her shoulder.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 해안가 좁은 길 (The two had been walking along it before 정인우 stopped); used as Narrow peripheral context indicating that the conversation has stopped along the route.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 정인우의 의심 어린 시선에 입술을 굳게 다물고 싸늘하게 식어버린 김지후(강수리영)의 굳은 표정.\n\nLOCATION (lock): A narrow rocky coastal path between the island village and the forest.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Finish the eye-level dolly-in on 김지후(강수리영)’s front three-quarter side, isolating her firmly closed lips and suddenly frozen expression while 정인우 remains beyond the frame on her eyeline. Her face occupies most of the image, with only a narrow trace of the coastal path retained at the edge before the camera is ready to ease past her shoulder.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) in the middle-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 해안가 좁은 길 (The two had been walking along it before 정인우 stopped); used as Narrow peripheral context indicating that the conversation has stopped along the route.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S56sh3__bgfirst_bg.png", "asset_id": "39cb3bb8-f9be-405b-a3b6-b95c7a1f9c98", "input_asset_ids": [ "f69df416-f47b-485d-8470-a711c431ca17", "f012a978-a0b0-49d3-8522-e8e0e14139ef" ] }, "S56sh3": { "input_fingerprint": "e90e9938e9663a6e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 정인우의 의심 어린 시선에 입술을 굳게 다물고 싸늘하게 식어버린 김지후(강수리영)의 굳은 표정.\n\nLOCATION (lock): A narrow rocky coastal path between the island village and the forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace, and Inwoo continues to have one arm in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 정인우의 의심 어린 시선에 입술을 굳게 다물고 싸늘하게 식어버린 김지후(강수리영)의 굳은 표정.\n\nLOCATION (lock): A narrow rocky coastal path between the island village and the forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace, and Inwoo continues to have one arm in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 정인우의 의심 어린 시선에 입술을 굳게 다물고 싸늘하게 식어버린 김지후(강수리영)의 굳은 표정.\n\nLOCATION (lock): A narrow rocky coastal path between the island village and the forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace, and Inwoo continues to have one arm in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S56sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S56sh3.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_섬_해안_바위길_446206.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "클로즈업 프레이밍을 통해 지문이 요구한 '굳게 다문 입술'과 '싸늘하게 식은 표정'을 매우 효과적으로 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "캐릭터와 배경은 잘 일치하나 프레이밍이 다소 넓어 지문의 핵심인 '표정'에 대한 집중도가 떨어집니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "클로즈업 프레이밍을 통해 지문이 요구한 '굳게 다문 입술'과 '싸늘하게 식은 표정'을 매우 효과적으로 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "캐릭터와 배경은 잘 일치하나 프레이밍이 다소 넓어 지문의 핵심인 '표정'에 대한 집중도가 떨어집니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "지시문의 핵심인 '굳은 표정'을 클로즈업으로 정확히 포착하여 프레이밍(우선순위 2)에서 확실한 우위를 점하며, 목걸이 줄 노출(우선순위 5)은 하위 항목이므로 감점의 영향이 적습니다." }, { "label": "A", "score": 5, "verdict_ko": "목걸이 미착용 설정(우선순위 5)은 잘 지켰으나, 표정을 강조해야 하는 텍스트 의도에 비해 샷의 프레이밍이 너무 넓어(우선순위 2) 몰입감이 떨어집니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지시문의 핵심인 '굳은 표정'을 클로즈업으로 정확히 포착하여 프레이밍(우선순위 2)에서 확실한 우위를 점하며, 목걸이 줄 노출(우선순위 5)은 하위 항목이므로 감점의 영향이 적습니다." }, { "label": "B", "score": 5, "verdict_ko": "목걸이 미착용 설정(우선순위 5)은 잘 지켰으나, 표정을 강조해야 하는 텍스트 의도에 비해 샷의 프레이밍이 너무 넓어(우선순위 2) 몰입감이 떨어집니다." } ] }, "combined": { "totals": { "A": 13, "B": 10 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 13, "B": 10 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "클로즈업 프레이밍을 통해 지문이 요구한 '굳게 다문 입술'과 '싸늘하게 식은 표정'을 매우 효과적으로 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "캐릭터와 배경은 잘 일치하나 프레이밍이 다소 넓어 지문의 핵심인 '표정'에 대한 집중도가 떨어집니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_섬_해안_바위길_446206.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S56sh3__bgfirst_bg.png", "bg_asset_id": "39cb3bb8-f9be-405b-a3b6-b95c7a1f9c98", "bg_record_key": "S56sh3::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "섬_해안_바위길", "groupbg_asset_id": "f012a978-a0b0-49d3-8522-e8e0e14139ef" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S56sh5": { "input_fingerprint": "a01f41d573dc05fe", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 멀리 떨어진 어선 위에서 숲으로 향하는 두 사람을 날카롭게 감시하는 황봉업(황 씨·민숙 집의 거구 남자)의 얼굴.\n\nLOCATION (lock): On the deck of a moored fishing boat, looking across the coast toward the rocky path. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear his cast while Hwang watches the pair from the boat; Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 4, "C": 2 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'얼굴' 샷치고는 앵글이 다소 넓으나, 지정된 돌담 해안가 배경과 인물의 복장, 일행을 감시하는 시선 처리를 가장 정확하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 얼굴은 잘 담아냈으나, 레퍼런스의 돌담길 대신 바다 위 돌길을 임의로 생성하여 장소(Location) 규칙을 위반했습니다." }, { "label": "C", "score": 2, "verdict_ko": "지정된 의상이 아닌 파란색 점퍼를 입고 있으며, 일행을 감시하지 않고 카메라를 응시하여 묘사 지침을 크게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S56sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S56sh3" } }, "S57sh1": { "input_fingerprint": "bceb40d075ffe741", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 이끼 낀 비석 표면에 '수.리.영'이라는 세 글자가 희미하게 새겨져 있는 클로즈업.\n\nLOCATION (lock): The mossy stone-marker area deep within the island forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The marker retains the faintly carved name “Su Ri-yeong” and its rotting-meat offering. Jihu remains without her necklace, and Inwoo's arm remains cast.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "'수.리.영'이라는 세 글자의 형태가 가장 정확하며, 클로즈업 샷과 고기 제물 등 프롬프트의 지시를 충실히 따랐습니다." }, { "label": "A", "score": 5, "verdict_ko": "전반적인 연출은 좋으나 '리' 자의 형태가 다소 왜곡되어 가독성이 떨어집니다." }, { "label": "B", "score": 4, "verdict_ko": "가운데 글자가 한자나 다른 기호처럼 형성되어 텍스트 구현에 실패했습니다." } ], "refs": [], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S57sh6": { "input_fingerprint": "370b51d4c6c9cb1e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 양손으로 거친 수풀을 양옆으로 밀어내며 숲속 깊은 곳을 향해 한 발을 내디딘 김지후(강수리영)의 단단한 뒷모습.\n\nLOCATION (lock): A dense, overgrown route leading away from the named forest markers. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same mossy markers, dense pine undergrowth, rotten offerings, and subdued daylight. Exclude the solitary earlier exploration and add the woman leading her companion deeper through the brush.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace while leading Inwoo through the forest; Inwoo's injured arm remains in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 4, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "샷 텍스트에 명시된 '양손으로 수풀을 밀어내는' 뒷모습과 발을 내디딘 동작을 가장 정확하게 구현했습니다(머리 스타일의 차이는 경미한 감점 요소)." }, { "label": "C", "score": 4, "verdict_ko": "양손으로 수풀을 밀어내는 동작이 정확히 표현되지 않았으며, 텍스트에 없는 인물이 배경에 등장해 지침을 위반했습니다." }, { "label": "A", "score": 3, "verdict_ko": "수풀을 밀어내는 대신 비석에 손을 얹고 있으며, 텍스트에 없는 인물이 정면에 크게 등장해 핵심 연출 지시를 어겼습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S57sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S57sh1" } }, "S58sh2::bgfirst_bg": { "input_fingerprint": "71346156e3a7707f", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 축사를 쳐다보며 과거의 공포가 떠오른 듯 한 손으로 가슴을 부여잡은 채 딱딱하게 굳어버린 김지후(강수리영)의 상체.\n\nLOCATION (lock): The empty passage through the forest goat enclosure in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-torso height and slightly off 김지후(강수리영)’s frontal axis, hold a static medium-close composition as her halted body occupies the right half and the empty enclosure remains along her off-screen eyeline to the left. Her hand clutches her chest at the center of the frame while her rigid shoulders and fixed gaze make the remembered terror readable without turning the image into a flashback.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 김지후(강수리영) in the middle-right of the frame, midground, looks toward empty goat enclosure; empty goat enclosure in the middle-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 텅 빈 염소 축사 (Empty; the black goat is absent) — The enclosure side facing the passing route is visible beside her eyeline; used as Fear trigger positioned along 김지후’s rigid eyeline; 축사 옆 길 (김지후와 정인우 are passing along it); used as Route context linking the present passage to 김지후’s previous experience.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 축사를 쳐다보며 과거의 공포가 떠오른 듯 한 손으로 가슴을 부여잡은 채 딱딱하게 굳어버린 김지후(강수리영)의 상체.\n\nLOCATION (lock): The empty passage through the forest goat enclosure in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-torso height and slightly off 김지후(강수리영)’s frontal axis, hold a static medium-close composition as her halted body occupies the right half and the empty enclosure remains along her off-screen eyeline to the left. Her hand clutches her chest at the center of the frame while her rigid shoulders and fixed gaze make the remembered terror readable without turning the image into a flashback.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 김지후(강수리영) in the middle-right of the frame, midground, looks toward empty goat enclosure; empty goat enclosure in the middle-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 텅 빈 염소 축사 (Empty; the black goat is absent) — The enclosure side facing the passing route is visible beside her eyeline; used as Fear trigger positioned along 김지후’s rigid eyeline; 축사 옆 길 (김지후와 정인우 are passing along it); used as Route context linking the present passage to 김지후’s previous experience.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S58sh2__bgfirst_bg.png", "asset_id": "536b7861-df9a-4dab-8402-c2ef45c444f2", "input_asset_ids": [ "b4edf495-d808-4e99-9dfa-5daa28468674", "b1ef5a8e-92a6-4bc1-89ca-de320c8d4a75" ] }, "S58sh2": { "input_fingerprint": "77b9d5dd45d884f2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 축사를 쳐다보며 과거의 공포가 떠오른 듯 한 손으로 가슴을 부여잡은 채 딱딱하게 굳어버린 김지후(강수리영)의 상체.\n\nLOCATION (lock): The empty passage through the forest goat enclosure in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The goat enclosure remains empty of the black goats. Jihu remains without her necklace, and Inwoo continues wearing his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 축사를 쳐다보며 과거의 공포가 떠오른 듯 한 손으로 가슴을 부여잡은 채 딱딱하게 굳어버린 김지후(강수리영)의 상체.\n\nLOCATION (lock): The empty passage through the forest goat enclosure in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The goat enclosure remains empty of the black goats. Jihu remains without her necklace, and Inwoo continues wearing his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 축사를 쳐다보며 과거의 공포가 떠오른 듯 한 손으로 가슴을 부여잡은 채 딱딱하게 굳어버린 김지후(강수리영)의 상체.\n\nLOCATION (lock): The empty passage through the forest goat enclosure in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The goat enclosure remains empty of the black goats. Jihu remains without her necklace, and Inwoo continues wearing his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S58sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S58sh2.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_숲속_염소_축사_f6a67e.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "샷 텍스트에 지정된 상체 프레이밍, 한 손으로 가슴을 부여잡은 자세, 빈 축사 배경 및 단독 등장 조건을 정확히 충족했습니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 언급되지 않은 인물(깁스를 한 소년)을 임의로 추가한 것은 심각한 규칙 위반입니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "샷 텍스트에 지정된 상체 프레이밍, 한 손으로 가슴을 부여잡은 자세, 빈 축사 배경 및 단독 등장 조건을 정확히 충족했습니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 언급되지 않은 인물(깁스를 한 소년)을 임의로 추가한 것은 심각한 규칙 위반입니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "샷 텍스트에 명시된 대로 김지후 1인만 등장하여 가슴을 부여잡은 상체 샷을 정확히 연출했으며, 레퍼런스의 장소와 인물 디테일도 충실히 반영했습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 텍스트에 언급되지 않은 인물(깁스를 한 남성)을 화면에 임의로 추가하여 엄격한 인물 제한 지시를 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "샷 텍스트에 명시된 대로 김지후 1인만 등장하여 가슴을 부여잡은 상체 샷을 정확히 연출했으며, 레퍼런스의 장소와 인물 디테일도 충실히 반영했습니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 언급되지 않은 인물(깁스를 한 남성)을 화면에 임의로 추가하여 엄격한 인물 제한 지시를 위반했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 15 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "샷 텍스트에 지정된 상체 프레이밍, 한 손으로 가슴을 부여잡은 자세, 빈 축사 배경 및 단독 등장 조건을 정확히 충족했습니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 언급되지 않은 인물(깁스를 한 소년)을 임의로 추가한 것은 심각한 규칙 위반입니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_숲속_염소_축사_f6a67e.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S58sh2__bgfirst_bg.png", "bg_asset_id": "536b7861-df9a-4dab-8402-c2ef45c444f2", "bg_record_key": "S58sh2::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "숲속_염소_축사", "groupbg_asset_id": "b1ef5a8e-92a6-4bc1-89ca-de320c8d4a75" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S59sh7": { "input_fingerprint": "d8ea241e7160ae96", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 열린 포댓자루 틈 사이로 피범벅이 된 사람의 팔 한 짝이 바닥으로 툭 떨어진 끔찍한 찰나.\n\nLOCATION (lock): The storage corner inside the daylight slaughterhouse, beside blood-soaked sacks. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): A severed, blood-covered human arm protrudes from the opened sack and drops onto the slaughterhouse floor, detached from the dismembered torso and other limbs packed inside. Jihu remains without her necklace, and Inwoo's arm remains cast.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 6, "B": 5, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 6, "verdict_ko": "바닥에 떨어진 팔의 구도는 가장 자연스러우나, 필수 지시사항인 '피범벅이 된 맨살' 대신 옷과 장갑을 묘사해 감점되었습니다." }, { "label": "B", "score": 5, "verdict_ko": "자루 밖으로 나온 팔을 연출했으나 형태가 다소 뻣뻣하며, 역시 피범벅 묘사가 누락되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "허공에 뻣뻣하게 떠 있는 빈 고무장갑을 그려 물리적으로 불가능한 연출(하드 위반)을 범했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B02.png" } ], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S59sh11::bgfirst_bg": { "input_fingerprint": "57ec6ce311676390", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 창고 구석 어둠 속에 웅크린 채 양손으로 입을 꽉 틀어막은 김지후(강수리영)와 정인우의 굳은 얼굴.\n\nLOCATION (lock): A cramped inner storeroom within the slaughterhouse, packed with sacks containing dismembered remains.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From crouched head height in the storage corner, hold a close, low eye-level two-shot about one meter in front of 김지후(강수리영) and 정인우, with their compressed bodies occupying opposite halves of the frame. Keep the blood-marked sacks soft but recognizable along the lower foreground while both faces remain sharply held in the same shallow plane, preserving the track’s settled concealment position before the pan toward the door.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 피 묻은 포대자루들 (속이 차 있고 검붉은 핏자국이 배어남) — The bulging sides nearest the hiding pair face the camera; used as Soft lower-foreground obstruction that compresses the hiding pair into the corner; 창고 구석 (어둠 속); used as Dark spatial boundary enclosing the pair behind and to both sides; 창고 문 (닫혀 있음) — The interior-facing side is angled away from the camera position; used as Off-axis destination for the pair's fixed attention and the coming pan.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 창고 구석 어둠 속에 웅크린 채 양손으로 입을 꽉 틀어막은 김지후(강수리영)와 정인우의 굳은 얼굴.\n\nLOCATION (lock): A cramped inner storeroom within the slaughterhouse, packed with sacks containing dismembered remains.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From crouched head height in the storage corner, hold a close, low eye-level two-shot about one meter in front of 김지후(강수리영) and 정인우, with their compressed bodies occupying opposite halves of the frame. Keep the blood-marked sacks soft but recognizable along the lower foreground while both faces remain sharply held in the same shallow plane, preserving the track’s settled concealment position before the pan toward the door.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 피 묻은 포대자루들 (속이 차 있고 검붉은 핏자국이 배어남) — The bulging sides nearest the hiding pair face the camera; used as Soft lower-foreground obstruction that compresses the hiding pair into the corner; 창고 구석 (어둠 속); used as Dark spatial boundary enclosing the pair behind and to both sides; 창고 문 (닫혀 있음) — The interior-facing side is angled away from the camera position; used as Off-axis destination for the pair's fixed attention and the coming pan.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S59sh11__bgfirst_bg.png", "asset_id": "6c7da4c9-d703-441d-8722-62995f5b84aa", "input_asset_ids": [ "9f31835c-1bad-489e-9a0a-54e9e9907b0c", "0d7fbf8f-f874-4d7d-ab18-bbfd382eb1bb" ] }, "S59sh11": { "input_fingerprint": "698e74eca022dc97", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 창고 구석 어둠 속에 웅크린 채 양손으로 입을 꽉 틀어막은 김지후(강수리영)와 정인우의 굳은 얼굴.\n\nLOCATION (lock): A cramped inner storeroom within the slaughterhouse, packed with sacks containing dismembered remains. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The storage space remains packed with blood-soaked sacks of dismembered human bodies, including the spilled limbs. Jihu remains without her necklace, and Inwoo still wears his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 창고 구석 어둠 속에 웅크린 채 양손으로 입을 꽉 틀어막은 김지후(강수리영)와 정인우의 굳은 얼굴.\n\nLOCATION (lock): A cramped inner storeroom within the slaughterhouse, packed with sacks containing dismembered remains. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The storage space remains packed with blood-soaked sacks of dismembered human bodies, including the spilled limbs. Jihu remains without her necklace, and Inwoo still wears his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 창고 구석 어둠 속에 웅크린 채 양손으로 입을 꽉 틀어막은 김지후(강수리영)와 정인우의 굳은 얼굴.\n\nLOCATION (lock): A cramped inner storeroom within the slaughterhouse, packed with sacks containing dismembered remains. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The storage space remains packed with blood-soaked sacks of dismembered human bodies, including the spilled limbs. Jihu remains without her necklace, and Inwoo still wears his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S59sh11__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S59sh11.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B02.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 복장과 소지품 상태(지후의 목걸이 없음, 인우의 깁스/붕대)를 잘 유지했으며, 양손으로 입을 막고 웅크린 포즈와 장소를 정확히 연출했습니다." }, { "label": "B", "score": 3, "verdict_ko": "지후가 목걸이를 착용하고 있어 명시된 지시사항을 위반했으며, 두 인물 모두 기준 이미지의 겉옷을 입고 있지 않아 복장 일관성이 떨어집니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 복장과 소지품 상태(지후의 목걸이 없음, 인우의 깁스/붕대)를 잘 유지했으며, 양손으로 입을 막고 웅크린 포즈와 장소를 정확히 연출했습니다." }, { "label": "B", "score": 3, "verdict_ko": "지후가 목걸이를 착용하고 있어 명시된 지시사항을 위반했으며, 두 인물 모두 기준 이미지의 겉옷을 입고 있지 않아 복장 일관성이 떨어집니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 대로 지후의 목걸이가 없고 의상이 레퍼런스와 잘 일치하며, 두 인물 모두 양손으로 입을 막고 웅크린 자세를 충실히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지후가 목걸이를 착용하지 않아야 한다는 지시를 위반했으며, 두 인물의 의상이 캐릭터 레퍼런스와 크게 다릅니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 대로 지후의 목걸이가 없고 의상이 레퍼런스와 잘 일치하며, 두 인물 모두 양손으로 입을 막고 웅크린 자세를 충실히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "지후가 목걸이를 착용하지 않아야 한다는 지시를 위반했으며, 두 인물의 의상이 캐릭터 레퍼런스와 크게 다릅니다." } ] }, "combined": { "totals": { "A": 14, "B": 7 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 7 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 복장과 소지품 상태(지후의 목걸이 없음, 인우의 깁스/붕대)를 잘 유지했으며, 양손으로 입을 막고 웅크린 포즈와 장소를 정확히 연출했습니다." }, { "label": "B", "score": 3, "verdict_ko": "지후가 목걸이를 착용하고 있어 명시된 지시사항을 위반했으며, 두 인물 모두 기준 이미지의 겉옷을 입고 있지 않아 복장 일관성이 떨어집니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B02.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S59sh11__bgfirst_bg.png", "bg_asset_id": "6c7da4c9-d703-441d-8722-62995f5b84aa", "bg_record_key": "S59sh11::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S59sh12": { "input_fingerprint": "2beb8bfe825d78cb", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 활짝 열린 창고 문 앞에 서서 내부를 빤히 들여다보는 고산들과 한국인 마을 청년들의 위협적인 실루엣.\n\nLOCATION (lock): The threshold between the slaughterhouse work floor and its inner sack storeroom. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rough interior, bloody floor, hanging metal tools, and daylight from the broken window. Exclude the earlier punishment scene and place the man and youths at the opened storage door.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The youths remove the blood-soaked sacks of human remains one by one, while Jihu and Inwoo remain hidden among the remaining sacks. Jihu lacks her necklace, and Inwoo's arm remains cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리); 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 5 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "정확한 리버스 앵글로 지시된 실루엣을 연출했으며, 기준 이미지의 슬라이딩 도어와 숨어있는 두 인물의 연속성까지 완벽히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "카메라가 작업장 내부에서 밝은 외부를 향하도록 잘못 배치되었고, 핵심인 실루엣 묘사와 숨겨진 캐릭터가 모두 누락되었습니다." }, { "label": "C", "score": 5, "verdict_ko": "실루엣 구도는 포착했으나, 확립된 슬라이딩 도어를 양문형 여닫이문으로 임의 변형하여 장소 일관성(Location Lock)을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S59sh11_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S59sh11" } }, "S60sh4": { "input_fingerprint": "bd75ddfac3440eb2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 바다 쪽을 향해 고개를 돌린 채 두 눈이 커다랗게 휘둥그레진 김지후(강수리영)의 경악한 얼굴.\n\nLOCATION (lock): A precarious rocky shoreline path overlooking the water below. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's motionless body remains lodged between the rocks and seawater. Jihu remains without her necklace, and Inwoo continues wearing his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 5, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "샷 텍스트에 집중해 지후의 경악한 얼굴과 시선을 정확히 담아냈으며, 지시문에 없는 다른 인물을 프레임에서 올바르게 배제했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지후의 표정 연출은 무난하나, 샷 텍스트에 언급되지 않은 배경 인물들을 화면에 포함시켜 인물 제한 규칙을 위반했습니다." }, { "label": "C", "score": 4, "verdict_ko": "샷 텍스트에 없는 다른 인물들을 프레임에 노출시켜 인물 제한을 위반했으며, 배경의 공간적 배치와 구도가 텍스트의 의도를 흩트립니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S56sh5_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S56sh5" } }, "S60sh5": { "input_fingerprint": "d1a86e6530720a24", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 거친 파도가 갯바위에 부딪혀 하얀 물보라가 공중에 멈춘 순간, 바위 틈에 기괴한 자세로 널브러져 있는 김여상의 축 늘어진 시체 전경.\n\nLOCATION (lock): The wave-washed rocks directly below the island’s coastal path. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's limp corpse is wedged in a contorted position between the coastal rocks and the water, with the body lifted and lowered slightly by the striking waves; Jihu lacks her necklace, and Inwoo's injured arm remains cast.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 4, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "공중에 멈춘 물보라와 갯바위 위의 기괴한 시체 자세를 지문대로 잘 구현했으며, 신체 구조 묘사에 오류가 없습니다." }, { "label": "A", "score": 4, "verdict_ko": "해안 산책로 배경은 참조와 정확히 일치하나, 오른팔 부근에 세 번째 손이 생성되는 치명적인 해부학적 오류가 있습니다." }, { "label": "C", "score": 3, "verdict_ko": "바위 틈의 시체 전경이라는 구도는 무난하나, 왼손이 손가락 없는 뭉툭한 기형으로 묘사된 치명적 해부학적 오류가 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S60sh4_sel.png" } ], "critique_skipped": true, "ref_mode": "lane(map_marker): 스케치", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S60sh4" }, "lane_policy": "share_plan_prev_bgonly" }, "S61sh1": { "input_fingerprint": "d0401b7b9bcb339a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 붉은 노을이 깔린 선착장 바닥, 넓은 비닐 포대 위에서 가슴팍이 처참하게 찢긴 채 누워 있는 김여상의 시체.\n\nLOCATION (lock): A cleared patch of pier pavement used to examine a recovered body at sunset. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's waterlogged corpse has been moved onto a spread plastic sheet at the dock and is stretched face-up there, with both arms and legs slack and the deep claw-like chest wound exposed.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 6, "B": 4, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 6, "verdict_ko": "시신이 비닐 위에 팔다리를 늘어뜨린 채 정자세로 누운 형태는 잘 구현되었으나, 지시문이 요구한 처참하게 찢긴 가슴 상처가 묘사되지 않고 옷으로 덮여 있습니다." }, { "label": "B", "score": 4, "verdict_ko": "시체를 가방으로 완전히 덮어버려, 인물의 구체적인 형태와 노출된 가슴 상처를 보여주어야 하는 지침을 위반했습니다." }, { "label": "A", "score": 3, "verdict_ko": "시신의 형태를 가려버렸을 뿐만 아니라, 프롬프트에 없는 테이블과 텍스트가 적힌 증거 마커를 임의로 추가해 감점되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L19B04.png" } ], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S61sh11": { "input_fingerprint": "dd63aec5c28ed617", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 정인우의 턱밑까지 바짝 다가선 채 험악하게 인상을 쓴 구창모의 위압적인 상체.\n\nLOCATION (lock): Beside the gathered villagers and recovered body on the island pier. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's corpse remains on the plastic sheet nearby with its torn chest visible. Jihu remains without her necklace, and Inwoo still wears his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 5, "B": 3, "C": 7 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "구도와 표정은 지시문과 부합하나, 구창모의 경찰 제복이 평상복으로 변형되었고 시신의 훼손된 흉부 묘사가 누락되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "정인우의 어깨에 얹힌 깁스 한 팔이 해부학적으로 불가능한 구조를 띠고 있는 치명적인 오류가 있습니다." }, { "label": "C", "score": 7, "verdict_ko": "명시된 샷의 프레이밍을 잘 구현했으며, 경찰 제복, 깁스, 훼손된 시신 등 세부적인 유지 조건(carried state)을 가장 충실히 반영했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L19B04.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "플레이트+콘티+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S61sh1" } }, "S61sh15::bgfirst_bg": { "input_fingerprint": "5bc2a9a6b5e0b189", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 정인우를 향해 한 손을 꼿꼿이 들어 올린 채 단호하게 앞을 막아선 강영란(흰 머리 여자)의 굳건한 상체.\n\nLOCATION (lock): The pier approach to a docked police patrol boat at dusk.\n\nTIME OF DAY (lock): sunset.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From 정인우's side of the gathering, settle the pan into a firm eye-level medium frame centered on 강영란(흰 머리 여자)'s raised palm and unyielding upper body. Her hand occupies the central foreground without exceeding natural scale, while 정인우 remains checked at one edge behind the opposing line, making the blocked route readable within a single plane.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground, points to blocked route toward the boat; 정인우 in the middle-left of the frame, foreground; route toward the police boat in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 마을 청년들의 대치선 (정인우 앞을 막고 있음); used as Compressed human barrier behind the raised hand, reinforcing that passage is denied; 경찰 순찰정 (시체를 실어 나를 준비가 됨) — Its dock-facing side is visible beyond the gathering; used as Background destination that gives the stopping gesture narrative meaning; 선착장 통로 (마을 사람들에게 가로막힘); used as Open ground separating 정인우 from the boat and visually cut by 강영란's hand.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe main pier extends seaward from the village-side approach and terminates at an outer pier end surrounded by the adjacent sea. Its main deck provides a continuous route between the village and the outer end, while a vessel berth edge directly contacts the water along the boat-access side. At the landward end, the approach continues into the small island village. The fishing-gear warehouse and Banwol Restaurant face this pier-to-village route, with the warehouse doorway opening onto the landing-side escape path and the restaurant entrance addressing the same general frontage. Small village houses occupy the inland area around the approach, with woodland interwoven among and behind them. This arrangement allows views inward from the outer pier toward the village, outward from the village frontage toward the berth and sea, and lateral coverage between the warehouse, restaurant frontage, and main pier deck.\n- pier end: seaward terminal end of the island pier\n- pier: main island pier deck\n- berth: water-contacting vessel berth edge\n- water: sea adjoining the island pier\n- approach: village-to-pier approach\n- houses: small island village houses around the pier approach\n- woodland: woodland surrounding the small island village\n- warehouse: fishing-gear warehouse beside the pier landing\n- door: pier-facing fishing-gear warehouse doorway\n- restaurant: Banwol Restaurant at the village-side pier frontage\n- entrance: pier-visible entrance of Banwol Restaurant\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 정인우를 향해 한 손을 꼿꼿이 들어 올린 채 단호하게 앞을 막아선 강영란(흰 머리 여자)의 굳건한 상체.\n\nLOCATION (lock): The pier approach to a docked police patrol boat at dusk.\n\nTIME OF DAY (lock): sunset.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From 정인우's side of the gathering, settle the pan into a firm eye-level medium frame centered on 강영란(흰 머리 여자)'s raised palm and unyielding upper body. Her hand occupies the central foreground without exceeding natural scale, while 정인우 remains checked at one edge behind the opposing line, making the blocked route readable within a single plane.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground, points to blocked route toward the boat; 정인우 in the middle-left of the frame, foreground; route toward the police boat in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 마을 청년들의 대치선 (정인우 앞을 막고 있음); used as Compressed human barrier behind the raised hand, reinforcing that passage is denied; 경찰 순찰정 (시체를 실어 나를 준비가 됨) — Its dock-facing side is visible beyond the gathering; used as Background destination that gives the stopping gesture narrative meaning; 선착장 통로 (마을 사람들에게 가로막힘); used as Open ground separating 정인우 from the boat and visually cut by 강영란's hand.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe main pier extends seaward from the village-side approach and terminates at an outer pier end surrounded by the adjacent sea. Its main deck provides a continuous route between the village and the outer end, while a vessel berth edge directly contacts the water along the boat-access side. At the landward end, the approach continues into the small island village. The fishing-gear warehouse and Banwol Restaurant face this pier-to-village route, with the warehouse doorway opening onto the landing-side escape path and the restaurant entrance addressing the same general frontage. Small village houses occupy the inland area around the approach, with woodland interwoven among and behind them. This arrangement allows views inward from the outer pier toward the village, outward from the village frontage toward the berth and sea, and lateral coverage between the warehouse, restaurant frontage, and main pier deck.\n- pier end: seaward terminal end of the island pier\n- pier: main island pier deck\n- berth: water-contacting vessel berth edge\n- water: sea adjoining the island pier\n- approach: village-to-pier approach\n- houses: small island village houses around the pier approach\n- woodland: woodland surrounding the small island village\n- warehouse: fishing-gear warehouse beside the pier landing\n- door: pier-facing fishing-gear warehouse doorway\n- restaurant: Banwol Restaurant at the village-side pier frontage\n- entrance: pier-visible entrance of Banwol Restaurant\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S61sh15__bgfirst_bg.png", "asset_id": "7f54806d-f929-40ba-a2a7-80ea6ca44ba1", "input_asset_ids": [ "07a4df6f-77e8-45ac-b70b-b9e630d2a343" ] }, "S61sh15": { "input_fingerprint": "aa5a33b3666b1f18", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 정인우를 향해 한 손을 꼿꼿이 들어 올린 채 단호하게 앞을 막아선 강영란(흰 머리 여자)의 굳건한 상체.\n\nLOCATION (lock): The pier approach to a docked police patrol boat at dusk. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dock paving, corpse scene, clustered villagers, and sunset light. Exclude the local officer's confrontation and show the elder blocking the visiting man.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's corpse remains at the dock awaiting removal. Jihu remains without her necklace, and Inwoo's injured arm remains cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 정인우를 향해 한 손을 꼿꼿이 들어 올린 채 단호하게 앞을 막아선 강영란(흰 머리 여자)의 굳건한 상체.\n\nLOCATION (lock): The pier approach to a docked police patrol boat at dusk. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's corpse remains at the dock awaiting removal. Jihu remains without her necklace, and Inwoo's injured arm remains cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 정인우를 향해 한 손을 꼿꼿이 들어 올린 채 단호하게 앞을 막아선 강영란(흰 머리 여자)의 굳건한 상체.\n\nLOCATION (lock): The pier approach to a docked police patrol boat at dusk. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's corpse remains at the dock awaiting removal. Jihu remains without her necklace, and Inwoo's injured arm remains cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 정인우를 향해 한 손을 꼿꼿이 들어 올린 채 단호하게 앞을 막아선 강영란(흰 머리 여자)의 굳건한 상체.\n\nLOCATION (lock): The pier approach to a docked police patrol boat at dusk. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's corpse remains at the dock awaiting removal. Jihu remains without her necklace, and Inwoo's injured arm remains cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S61sh15__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S61sh15__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S61sh15__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "A": 7, "B": 4, "C": 5 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 굳건한 포즈와 배경을 잘 구현했으며, 레퍼런스의 한복과 두건 디테일을 정확히 유지해 가장 우수함." }, { "label": "B", "score": 4, "verdict_ko": "카메라 앵글은 상체에 더 가깝지만, 지정된 녹색 한복이 아닌 현대복으로 복장이 완전히 변경되어 감점됨." }, { "label": "C", "score": 5, "verdict_ko": "전반적인 샷 구성은 부합하나, 레퍼런스의 두건이 누락되고 치마가 바지로 변형되는 등 복장 일치도가 떨어짐." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S61sh15__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S61sh15__bgfirst_bg.png", "bg_asset_id": "7f54806d-f929-40ba-a2a7-80ea6ca44ba1", "bg_record_key": "S61sh15::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S61sh11" } }, "S62sh4::bgfirst_bg": { "input_fingerprint": "3322299e2f659224", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 김여상의 손목에서 번쩍이는 금속 시계를 절반쯤 빼낸 채 팽팽하게 힘을 주고 있는 구창모의 탐욕스러운 손아귀.\n\nLOCATION (lock): The open aft deck of a police patrol boat stopped on dark evening water.\n\nTIME OF DAY (lock): early evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At wrist height on the patrol-boat deck, hold a static insert inches from 김여상's slack forearm as 구창모's gripping hand strains the metal watch halfway free. Keep a narrow portion of the corpse's clothing and deck at the frame edges for scale, with the watch, wrist, and whitening grip forming the central diagonal rather than allowing the object alone to occupy the image.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 금속 시계 (손목에서 절반쯤 빠져 있음) — Its outer face and loosened band are angled toward the camera as it slides over the hand; used as Central evidence of theft, held between the corpse's wrist and 구창모's grip; 순찰정 갑판 (엔진을 끈 배 위); used as Thin lower-edge context fixing the action aboard the patrol boat; 김여상의 옷 (구창모가 뒤진 상태); used as Frame-edge evidence of the ongoing search.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 김여상의 손목에서 번쩍이는 금속 시계를 절반쯤 빼낸 채 팽팽하게 힘을 주고 있는 구창모의 탐욕스러운 손아귀.\n\nLOCATION (lock): The open aft deck of a police patrol boat stopped on dark evening water.\n\nTIME OF DAY (lock): early evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At wrist height on the patrol-boat deck, hold a static insert inches from 김여상's slack forearm as 구창모's gripping hand strains the metal watch halfway free. Keep a narrow portion of the corpse's clothing and deck at the frame edges for scale, with the watch, wrist, and whitening grip forming the central diagonal rather than allowing the object alone to occupy the image.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 금속 시계 (손목에서 절반쯤 빠져 있음) — Its outer face and loosened band are angled toward the camera as it slides over the hand; used as Central evidence of theft, held between the corpse's wrist and 구창모's grip; 순찰정 갑판 (엔진을 끈 배 위); used as Thin lower-edge context fixing the action aboard the patrol boat; 김여상의 옷 (구창모가 뒤진 상태); used as Frame-edge evidence of the ongoing search.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S62sh4__bgfirst_bg.png", "asset_id": "9b8021ca-3495-49ee-9384-b9099569d27d", "input_asset_ids": [ "27bc1894-fc46-4c44-97b3-dc91fa8ff411", "952ffc2c-6ca7-4738-a177-f23fca6c2548" ] }, "S62sh4": { "input_fingerprint": "4d6ad53fd36b05b1", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): early evening.\n\nSHOT TEXT (authoritative, Korean): 김여상의 손목에서 번쩍이는 금속 시계를 절반쯤 빼낸 채 팽팽하게 힘을 주고 있는 구창모의 탐욕스러운 손아귀.\n\nLOCATION (lock): The open aft deck of a police patrol boat stopped on dark evening water. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): His corpse is stretched motionless on the patrol boat's deck, with the torso supported by the deck and the arms resting slack while 구창모 handles one wrist and searches his clothing.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's corpse is stretched motionless on the patrol boat's deck, with the torso supported by the deck and the arms resting slack while Changmo handles one wrist and searches his clothing; the wet cash and ring are transferred to Changmo's pocket.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리); 김여상 (40대 중반 한국인 남성, 평범한 중년 얼굴, 단정한 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): early evening.\n\nSHOT TEXT (authoritative, Korean): 김여상의 손목에서 번쩍이는 금속 시계를 절반쯤 빼낸 채 팽팽하게 힘을 주고 있는 구창모의 탐욕스러운 손아귀.\n\nLOCATION (lock): The open aft deck of a police patrol boat stopped on dark evening water. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): His corpse is stretched motionless on the patrol boat's deck, with the torso supported by the deck and the arms resting slack while 구창모 handles one wrist and searches his clothing.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's corpse is stretched motionless on the patrol boat's deck, with the torso supported by the deck and the arms resting slack while Changmo handles one wrist and searches his clothing; the wet cash and ring are transferred to Changmo's pocket.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리); 김여상 (40대 중반 한국인 남성, 평범한 중년 얼굴, 단정한 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): early evening.\n\nSHOT TEXT (authoritative, Korean): 김여상의 손목에서 번쩍이는 금속 시계를 절반쯤 빼낸 채 팽팽하게 힘을 주고 있는 구창모의 탐욕스러운 손아귀.\n\nLOCATION (lock): The open aft deck of a police patrol boat stopped on dark evening water. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): His corpse is stretched motionless on the patrol boat's deck, with the torso supported by the deck and the arms resting slack while 구창모 handles one wrist and searches his clothing.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's corpse is stretched motionless on the patrol boat's deck, with the torso supported by the deck and the arms resting slack while Changmo handles one wrist and searches his clothing; the wet cash and ring are transferred to Changmo's pocket.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리); 김여상 (40대 중반 한국인 남성, 평범한 중년 얼굴, 단정한 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S62sh4__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S62sh4.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김여상: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L45B02.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김여상: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 강조한 손아귀와 시계를 빼내는 구체적인 동작을 정확한 클로즈업 샷으로 구현하여 최우선 기준을 충족했으나, 시신의 의상 색상이 레퍼런스와 다릅니다." }, { "label": "B", "score": 4, "verdict_ko": "캐릭터의 의상과 배경은 레퍼런스에 부합하나, 손아귀와 손목에 집중해야 할 지문의 묘사를 무시하고 불필요하게 넓은 풀샷으로 연출해 샷의 의도를 벗어났습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 강조한 손아귀와 시계를 빼내는 구체적인 동작을 정확한 클로즈업 샷으로 구현하여 최우선 기준을 충족했으나, 시신의 의상 색상이 레퍼런스와 다릅니다." }, { "label": "B", "score": 4, "verdict_ko": "캐릭터의 의상과 배경은 레퍼런스에 부합하나, 손아귀와 손목에 집중해야 할 지문의 묘사를 무시하고 불필요하게 넓은 풀샷으로 연출해 샷의 의도를 벗어났습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트가 묘사한 시계를 빼는 손아귀 중심의 클로즈업 대신 불필요하게 넓은 전신 화각을 보여주어 연출 지시를 크게 위반했습니다." }, { "label": "B", "score": 8, "verdict_ko": "시계를 절반쯤 빼내며 팽팽하게 힘을 주는 손아귀의 동작을 지시문이 요구한 정확한 클로즈업 구도로 훌륭하게 구현했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 3, "verdict_ko": "샷 텍스트가 묘사한 시계를 빼는 손아귀 중심의 클로즈업 대신 불필요하게 넓은 전신 화각을 보여주어 연출 지시를 크게 위반했습니다." }, { "label": "A", "score": 8, "verdict_ko": "시계를 절반쯤 빼내며 팽팽하게 힘을 주는 손아귀의 동작을 지시문이 요구한 정확한 클로즈업 구도로 훌륭하게 구현했습니다." } ] }, "combined": { "totals": { "A": 15, "B": 7 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 15, "B": 7 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 강조한 손아귀와 시계를 빼내는 구체적인 동작을 정확한 클로즈업 샷으로 구현하여 최우선 기준을 충족했으나, 시신의 의상 색상이 레퍼런스와 다릅니다." }, { "label": "B", "score": 4, "verdict_ko": "캐릭터의 의상과 배경은 레퍼런스에 부합하나, 손아귀와 손목에 집중해야 할 지문의 묘사를 무시하고 불필요하게 넓은 풀샷으로 연출해 샷의 의도를 벗어났습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L45B02.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김여상: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S62sh4__bgfirst_bg.png", "bg_asset_id": "9b8021ca-3495-49ee-9384-b9099569d27d", "bg_record_key": "S62sh4::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S62sh6": { "input_fingerprint": "f7fcbc9dd8b62a8a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): early evening.\n\nSHOT TEXT (authoritative, Korean): 나란히 맞닿아 출렁이는 두 배의 틈 사이로 김여상의 시체를 번쩍 들어 올린 채 굳건히 버티고 있는 황봉업(황 씨·민숙 집의 거구 남자)의 산 같은 뒷모습.\n\nLOCATION (lock): Across the narrow gap between a police patrol boat and a fishing boat rafted together at sea. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): His limp corpse is held horizontally above the gap between the two boats in 황봉업's arms, with the head, arms, and legs hanging without support as the body is transferred.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Yeosang's corpse is transferred from the patrol boat to Hwang's fishing boat, along with the clothing and remaining personal effects; Changmo now wears Yeosang's Rolex.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리); 김여상 (40대 중반 한국인 남성, 평범한 중년 얼굴, 단정한 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 3, "B": 4, "C": 7 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "들려있는 시신의 상하체가 분리되고 다리 방향이 어긋나는 등 물리적으로 불가능한 해부학적 오류가 발생한 치명적 위반입니다." }, { "label": "B", "score": 4, "verdict_ko": "숏 텍스트에 명시되지 않은 제3의 인물(우측에 시계를 찬 팔)이 프레임에 등장하여 등장인물 제한 규칙을 위반했습니다." }, { "label": "C", "score": 7, "verdict_ko": "두 배 사이에서 시신을 들고 버티는 뒷모습의 구도를 정확히 구현했으며, 늘어진 시신의 자세와 이전 숏의 복장 유지도 훌륭합니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S62sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 김여상: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S62sh4" }, "lane_policy": "ab_select_bypass:prev" }, "S62sh9": { "input_fingerprint": "4c0af43d1564d997", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): early evening.\n\nSHOT TEXT (authoritative, Korean): 서로 반대 방향의 캄캄한 밤바다를 향해 선미 뒤로 하얀 항적을 길게 남기며 물살을 가르고 있는 경찰 순찰정과 낚시 어선의 고독한 조감도.\n\nLOCATION (lock): Open night sea where the two boats depart in opposite directions, leaving long wakes. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hwang's fishing boat carries Yeosang's corpse back toward Geumwoldo, while Changmo departs wearing Yeosang's Rolex and carrying the stolen valuables.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 4, "B": 7, "C": 5 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "두 배가 V자 형태로 나아가고 있어 '서로 반대 방향'이라는 핵심 지시를 명확히 구현하지 못했습니다." }, { "label": "B", "score": 7, "verdict_ko": "두 배가 서로 반대 방향으로 멀어지며 긴 항적을 남기는 조감도 구도와 밤바다의 분위기를 가장 잘 표현했습니다." }, { "label": "C", "score": 5, "verdict_ko": "반대 방향으로 이동하고 있으나, 두 배의 항적이 비현실적으로 하나의 일직선으로 이어져 있어 다소 어색합니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S62sh6_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S62sh6" }, "lane_policy": "share_plan_prev_bgonly" }, "S63sh17::bgfirst_bg": { "input_fingerprint": "4f08e0d170e67e74", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 입안 가득 붉은 생고기를 물고 있는 김지후(강수리영)의 핏물 묻은 입가.\n\nLOCATION (lock): At the raw-meat dinner table inside the restaurant hall under nighttime interior light.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the table side at mouth height, complete the dolly-in on an extreme close three-quarter view of 김지후(강수리영), isolating her blood-marked mouth and the raw meat between her teeth while excluding the rest of the table. Her lower face occupies most of the frame, with shallow focus holding the wet bite and smeared lips as the only readable action.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) mouth and bite in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: raw meat (raw and blood-red); used as Held in the central mouth-level focal plane as the grotesque narrative detail.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 입안 가득 붉은 생고기를 물고 있는 김지후(강수리영)의 핏물 묻은 입가.\n\nLOCATION (lock): At the raw-meat dinner table inside the restaurant hall under nighttime interior light.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the table side at mouth height, complete the dolly-in on an extreme close three-quarter view of 김지후(강수리영), isolating her blood-marked mouth and the raw meat between her teeth while excluding the rest of the table. Her lower face occupies most of the frame, with shallow focus holding the wet bite and smeared lips as the only readable action.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) mouth and bite in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: raw meat (raw and blood-red); used as Held in the central mouth-level focal plane as the grotesque narrative detail.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S63sh17__bgfirst_bg.png", "asset_id": "1a7d58fc-1c65-4345-9b18-dafb1ac697cd", "input_asset_ids": [ "62a7ec6a-d597-409d-bec5-e834f480e23b", "7708314a-bf99-4888-bd2e-f885cf704b49" ] }, "S63sh17": { "input_fingerprint": "e87f97dd584f94e1", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 입안 가득 붉은 생고기를 물고 있는 김지후(강수리영)의 핏물 묻은 입가.\n\nLOCATION (lock): At the raw-meat dinner table inside the restaurant hall under nighttime interior light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace and has fresh blood and raw meat smeared around her mouth and hands. Inwoo continues to wear his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 입안 가득 붉은 생고기를 물고 있는 김지후(강수리영)의 핏물 묻은 입가.\n\nLOCATION (lock): At the raw-meat dinner table inside the restaurant hall under nighttime interior light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace and has fresh blood and raw meat smeared around her mouth and hands. Inwoo continues to wear his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 입안 가득 붉은 생고기를 물고 있는 김지후(강수리영)의 핏물 묻은 입가.\n\nLOCATION (lock): At the raw-meat dinner table inside the restaurant hall under nighttime interior light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace and has fresh blood and raw meat smeared around her mouth and hands. Inwoo continues to wear his cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S63sh17__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S63sh17.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 붉은 생고기: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 붉은 생고기: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "야간 실내 조명과 장소 연출이 자연스러우며, 입에 생고기를 물고 있는 캐릭터의 디테일이 지시문에 부합합니다." }, { "label": "A", "score": 3, "verdict_ko": "고기를 쥐고 있는 손가락 구조가 고기와 뭉개진 물리적 해부학 오류(치명적 위반)가 발생했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "야간 실내 조명과 장소 연출이 자연스러우며, 입에 생고기를 물고 있는 캐릭터의 디테일이 지시문에 부합합니다." }, { "label": "A", "score": 3, "verdict_ko": "고기를 쥐고 있는 손가락 구조가 고기와 뭉개진 물리적 해부학 오류(치명적 위반)가 발생했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 야간 실내 조명 속에서 생고기를 물고 있는 핏물 묻은 입가라는 샷 텍스트를 사실적으로 잘 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "배경이 레퍼런스 이미지를 그대로 잘라 붙인 듯한 합성(콜라주) 형태를 띠며, 고기를 쥔 손의 해부학적 구조가 어색합니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 야간 실내 조명 속에서 생고기를 물고 있는 핏물 묻은 입가라는 샷 텍스트를 사실적으로 잘 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "배경이 레퍼런스 이미지를 그대로 잘라 붙인 듯한 합성(콜라주) 형태를 띠며, 고기를 쥔 손의 해부학적 구조가 어색합니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "야간 실내 조명과 장소 연출이 자연스러우며, 입에 생고기를 물고 있는 캐릭터의 디테일이 지시문에 부합합니다." }, { "label": "A", "score": 3, "verdict_ko": "고기를 쥐고 있는 손가락 구조가 고기와 뭉개진 물리적 해부학 오류(치명적 위반)가 발생했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B01.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 붉은 생고기: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S63sh17__bgfirst_bg.png", "bg_asset_id": "1a7d58fc-1c65-4345-9b18-dafb1ac697cd", "bg_record_key": "S63sh17::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S63sh28": { "input_fingerprint": "017853491877df9a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 피투성이 손을 목격하고 경악하여 입을 크게 벌린 채 굳어 있는 김지후(강수리영)의 일그러진 얼굴.\n\nLOCATION (lock): The dining-table area inside the restaurant, surrounded by platters of raw meat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same wooden dining room, dark windows, table location, and low night lighting. Exclude the empty-room composition and retain the blood-covered dinner table while the woman recoils from her hands.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu's mouth and both hands remain heavily smeared with raw blood after eating the meat, and her necklace is still missing. Inwoo's injured arm remains cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 피 묻은 식탁과 고기 쟁반을 정확히 유지했으며, 피투성이 손을 보고 경악하여 굳은 표정을 훌륭하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "경악하여 일그러진 얼굴과 피 묻은 손의 묘사는 우수하나, 프롬프트에서 유지하라고 명시한 '피 묻은 저녁 식탁'이 프레임에서 배제되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "필수 요소인 전경의 피 묻은 식탁과 고기 쟁반이 생략되었고, 손의 형태가 다소 어색하며 앵글이 지시된 구도와 어긋납니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S63sh17_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S63sh17" } }, "S63sh30": { "input_fingerprint": "8e78ac41f8acb90b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혼란스러운 두 사람을 내려다보며 입꼬리가 올라간 기괴한 미소를 머금고 있는 강영란(흰 머리 여자).\n\nLOCATION (lock): Beside the dining table in the restaurant hall, with the adjoining room and kitchen nearby. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dim restaurant hall, wooden lattice windows, table placement, and nighttime light. Exclude the empty-room state and frame the elder watching the distressed pair with a satisfied smile.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains blood-smeared around the mouth and hands and still lacks her necklace; Inwoo remains beside her with one arm in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 5, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 앵글로 두 사람을 내려다보는 노파의 기괴한 미소를 정확히 연출했으며, 식탁 위의 생고기 등 이전 샷의 연속성을 잘 유지했습니다." }, { "label": "A", "score": 5, "verdict_ko": "노파의 시선과 표정 연출은 적절하나, 이전 샷에서 명확히 설정된 식탁 위 생고기들이 모두 사라져 환경 연속성(Priority 3)을 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "노파가 인물들이 아닌 카메라를 응시하여 연출 지시를 어겼으며, 이전 샷의 카메라 구도와 인물 자세를 그대로 복사하지 말라는 금지 사항을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S63sh28_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S63sh28" } }, "S64sh2::bgfirst_bg": { "input_fingerprint": "73755589a7bec72d", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 칠흑 같은 숲길 오르막에서 한 발을 앞서 내디딘 채 걷는 도중인 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): A steep, unlit ascent through the maritime-pine forest beneath the nearly full moon.\n\nTIME OF DAY (lock): night, near-full moon, starless.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track at standing eye level several paces behind 강영란(흰 머리 여자), aligned directly with her uphill route, and hold a wide rear composition as one foot advances. Her erect back occupies the central vertical axis while the climbing path and forest retain enough space to establish her fearless progress before the camera lowers in the next stage.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground, moves toward upper stretch of the uphill path; upper stretch of the uphill path in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: uphill forest path (dark at night); used as Forms the central route and carries the forward tracking axis; forest (dark at night); used as Encloses both sides of the path and narrows the visible route.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 칠흑 같은 숲길 오르막에서 한 발을 앞서 내디딘 채 걷는 도중인 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): A steep, unlit ascent through the maritime-pine forest beneath the nearly full moon.\n\nTIME OF DAY (lock): night, near-full moon, starless.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track at standing eye level several paces behind 강영란(흰 머리 여자), aligned directly with her uphill route, and hold a wide rear composition as one foot advances. Her erect back occupies the central vertical axis while the climbing path and forest retain enough space to establish her fearless progress before the camera lowers in the next stage.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground, moves toward upper stretch of the uphill path; upper stretch of the uphill path in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: uphill forest path (dark at night); used as Forms the central route and carries the forward tracking axis; forest (dark at night); used as Encloses both sides of the path and narrows the visible route.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S64sh2__bgfirst_bg.png", "asset_id": "5e32a015-001c-44f3-be4e-1503506a0fae", "input_asset_ids": [ "10047cab-1ea1-4e24-8fb8-972507aee326", "dded1895-8314-4902-9d6c-855ef980a51c" ] }, "S64sh2": { "input_fingerprint": "dad55859072920b0", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, near-full moon, starless.\n\nSHOT TEXT (authoritative, Korean): 칠흑 같은 숲길 오르막에서 한 발을 앞서 내디딘 채 걷는 도중인 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): A steep, unlit ascent through the maritime-pine forest beneath the nearly full moon. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe as she travels toward the ritual site.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, near-full moon, starless.\n\nSHOT TEXT (authoritative, Korean): 칠흑 같은 숲길 오르막에서 한 발을 앞서 내디딘 채 걷는 도중인 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): A steep, unlit ascent through the maritime-pine forest beneath the nearly full moon. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe as she travels toward the ritual site.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, near-full moon, starless.\n\nSHOT TEXT (authoritative, Korean): 칠흑 같은 숲길 오르막에서 한 발을 앞서 내디딘 채 걷는 도중인 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): A steep, unlit ascent through the maritime-pine forest beneath the nearly full moon. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe as she travels toward the ritual site.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S64sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S64sh2.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "프롬프트가 명시한 '뒷모습'과 오르막을 오르는 방향을 완전히 위반하여 정면을 보여주며, 얼굴을 알 수 없는 형태로 뭉개버려 심각한 오류가 있습니다." }, { "label": "B", "score": 8, "verdict_ko": "요구된 '뒷모습' 앵글과 한 발을 앞서 내디디며 오르막을 오르는 동작, 붉은 무당복과 틀어 올린 흰 머리까지 모든 지시사항을 매우 정확하게 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "프롬프트가 명시한 '뒷모습'과 오르막을 오르는 방향을 완전히 위반하여 정면을 보여주며, 얼굴을 알 수 없는 형태로 뭉개버려 심각한 오류가 있습니다." }, { "label": "B", "score": 8, "verdict_ko": "요구된 '뒷모습' 앵글과 한 발을 앞서 내디디며 오르막을 오르는 동작, 붉은 무당복과 틀어 올린 흰 머리까지 모든 지시사항을 매우 정확하게 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "오르막길을 걷는 꼿꼿한 뒷모습과 의상, 밤의 숲길 배경을 지문과 레퍼런스에 맞게 충실히 구현했습니다." }, { "label": "B", "score": 2, "verdict_ko": "뒷모습이라는 지문과 정반대로 정면을 향하고 있으며, 얼굴이 기괴한 형태로 묘사되어 인체 표현 지침을 심각하게 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "오르막길을 걷는 꼿꼿한 뒷모습과 의상, 밤의 숲길 배경을 지문과 레퍼런스에 맞게 충실히 구현했습니다." }, { "label": "A", "score": 2, "verdict_ko": "뒷모습이라는 지문과 정반대로 정면을 향하고 있으며, 얼굴이 기괴한 형태로 묘사되어 인체 표현 지침을 심각하게 위반했습니다." } ] }, "combined": { "totals": { "A": 4, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 4, "B": 16 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "프롬프트가 명시한 '뒷모습'과 오르막을 오르는 방향을 완전히 위반하여 정면을 보여주며, 얼굴을 알 수 없는 형태로 뭉개버려 심각한 오류가 있습니다." }, { "label": "B", "score": 8, "verdict_ko": "요구된 '뒷모습' 앵글과 한 발을 앞서 내디디며 오르막을 오르는 동작, 붉은 무당복과 틀어 올린 흰 머리까지 모든 지시사항을 매우 정확하게 구현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S64sh2__bgfirst_bg.png", "bg_asset_id": "5e32a015-001c-44f3-be4e-1503506a0fae", "bg_record_key": "S64sh2::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "해송_숲길", "groupbg_asset_id": "dded1895-8314-4902-9d6c-855ef980a51c" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S65sh1::bgfirst_bg": { "input_fingerprint": "1814bb539a1dbb59", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 어두운 식당 밖 골목에 흩어져 서서 식당을 향해 시선을 고정한 한국인 마을 청년들의 위협적인 전경.\n\nLOCATION (lock): The dark village lane immediately outside the restaurant entrance.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level near one end of the alley, hold a static wide view down the dispersed line of Korean village youths toward the restaurant. Place the nearest youth on one outer third and the others at irregular midground intervals, all sharing the restaurant as their eyeline while negative space between them makes the surveillance feel coordinated rather than casual.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 마을 청년#1 in the lower-left of the frame, foreground, looks toward restaurant; restaurant in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: restaurant exterior (under observation at night) — The side facing the alley is visible to camera; used as Shared endpoint of every youth's eyeline at the far side of the alley; alley (occupied by residents and youths); used as Provides the depth axis along which the separated watchers are distributed.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe Banwol Restaurant main building is the organizing mass of the site. Its front entrance opens onto the restaurant-front alley, which joins the village-road approach aligned toward the entrance. From the restaurant side, the village road also continues uphill until it reaches the forest and mountain threshold. Behind the main building, the back entrance opens directly into the restaurant rear yard. The yard contains a cultivated garden plot and the separate storage shed, with open circulation linking the back entrance, garden edge, and shed door. The shed door faces into the yard so cameras can work from the yard or directly at the shed threshold. A perimeter boundary defines the rear-yard extent while preserving its connections to the building and village-road side.\n- building: Banwol Restaurant main building\n- entrance: Restaurant front entrance\n- alley: Restaurant-front alley\n- approach: Restaurant-approach village road\n- yard: Restaurant rear yard\n- garden: Cultivated rear-yard garden plot\n- entrance: Restaurant back entrance\n- boundary: Rear-yard perimeter boundary\n- shed: Rear-yard storage shed\n- door: Rear-yard storage shed door\n- road: Uphill village-road continuation\n- boundary: Forest and mountain threshold\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 어두운 식당 밖 골목에 흩어져 서서 식당을 향해 시선을 고정한 한국인 마을 청년들의 위협적인 전경.\n\nLOCATION (lock): The dark village lane immediately outside the restaurant entrance.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level near one end of the alley, hold a static wide view down the dispersed line of Korean village youths toward the restaurant. Place the nearest youth on one outer third and the others at irregular midground intervals, all sharing the restaurant as their eyeline while negative space between them makes the surveillance feel coordinated rather than casual.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 마을 청년#1 in the lower-left of the frame, foreground, looks toward restaurant; restaurant in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: restaurant exterior (under observation at night) — The side facing the alley is visible to camera; used as Shared endpoint of every youth's eyeline at the far side of the alley; alley (occupied by residents and youths); used as Provides the depth axis along which the separated watchers are distributed.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe Banwol Restaurant main building is the organizing mass of the site. Its front entrance opens onto the restaurant-front alley, which joins the village-road approach aligned toward the entrance. From the restaurant side, the village road also continues uphill until it reaches the forest and mountain threshold. Behind the main building, the back entrance opens directly into the restaurant rear yard. The yard contains a cultivated garden plot and the separate storage shed, with open circulation linking the back entrance, garden edge, and shed door. The shed door faces into the yard so cameras can work from the yard or directly at the shed threshold. A perimeter boundary defines the rear-yard extent while preserving its connections to the building and village-road side.\n- building: Banwol Restaurant main building\n- entrance: Restaurant front entrance\n- alley: Restaurant-front alley\n- approach: Restaurant-approach village road\n- yard: Restaurant rear yard\n- garden: Cultivated rear-yard garden plot\n- entrance: Restaurant back entrance\n- boundary: Rear-yard perimeter boundary\n- shed: Rear-yard storage shed\n- door: Rear-yard storage shed door\n- road: Uphill village-road continuation\n- boundary: Forest and mountain threshold\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S65sh1__bgfirst_bg.png", "asset_id": "f6aeff03-4867-4c2e-8c67-b8b0892abd43", "input_asset_ids": [ "7cfa7361-af42-4c1b-bf5a-04ca99b0601d" ] }, "S65sh1": { "input_fingerprint": "2aefb1a1438b713d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 식당 밖 골목에 흩어져 서서 식당을 향해 시선을 고정한 한국인 마을 청년들의 위협적인 전경.\n\nLOCATION (lock): The dark village lane immediately outside the restaurant entrance. The shot takes place here — the attached STORYBOARD SKETCH fixes the staging, camera and figure placement of this exact place. No location photograph is attached — build the location itself strictly from the location text above and the shot text, inventing nothing beyond them.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The village watchers remain stationed around the restaurant, and the burned-blood residue from Jihu's meal remains on her unless cleaned offscreen without textual confirmation.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 식당 밖 골목에 흩어져 서서 식당을 향해 시선을 고정한 한국인 마을 청년들의 위협적인 전경.\n\nLOCATION (lock): The dark village lane immediately outside the restaurant entrance. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The village watchers remain stationed around the restaurant, and the burned-blood residue from Jihu's meal remains on her unless cleaned offscreen without textual confirmation.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 식당 밖 골목에 흩어져 서서 식당을 향해 시선을 고정한 한국인 마을 청년들의 위협적인 전경.\n\nLOCATION (lock): The dark village lane immediately outside the restaurant entrance. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The village watchers remain stationed around the restaurant, and the burned-blood residue from Jihu's meal remains on her unless cleaned offscreen without textual confirmation.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 식당 밖 골목에 흩어져 서서 식당을 향해 시선을 고정한 한국인 마을 청년들의 위협적인 전경.\n\nLOCATION (lock): The dark village lane immediately outside the restaurant entrance. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The village watchers remain stationed around the restaurant, and the burned-blood residue from Jihu's meal remains on her unless cleaned offscreen without textual confirmation.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S65sh1__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S65sh1__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S65sh1__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "B": 9, "A": 6, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "스토리보드의 구도를 완벽히 재현했으며, 등장하는 모든 인물을 지정된 레퍼런스(마을 청년#1)의 외모와 복장으로 정확히 통일했습니다." }, { "label": "A", "score": 6, "verdict_ko": "구도와 시선 방향은 스토리보드와 일치하지만, 지정된 인물 레퍼런스의 복장(파란색 후드티)이 한 명에게만 적용되고 나머지는 다른 옷을 입어 감점되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "허용되지 않은 인물(긴 머리의 여성)이 등장하며, 지정된 레퍼런스의 복장 및 외모와 일치하지 않는 인물들이 포함되어 심각한 오류가 발생했습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S65sh1__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S65sh1__bgfirst_bg.png", "bg_asset_id": "f6aeff03-4867-4c2e-8c67-b8b0892abd43", "bg_record_key": "S65sh1::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+엔티티", "share_plan": { "ref_plan": "background" } }, "S66sh1": { "input_fingerprint": "161b88ec74c3f173", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moon obscured by smoke.\n\nSHOT TEXT (authoritative, Korean): 식당 뒤뜰, 시뻘건 불길에 휩싸여 활활 타오르고 있는 낡은 창고의 압도적인 전경.\n\nLOCATION (lock): The restaurant’s rear yard around a wooden shed fully engulfed in flames. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rear shed is fully ablaze, with thick black smoke rising above it; the fire persists while the villagers attempt to extinguish it.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 8, "B": 6, "A": 4 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "LOCATION의 주변 환경을 정확히 유지하면서, 불타는 창고만 STRUCTURE LOOK의 골판지 금속 재질과 회색 문 형태에 맞게 완벽히 교체함." }, { "label": "B", "score": 6, "verdict_ko": "창고 외의 주변부는 LOCATION을 따라야 한다는 규칙을 어기고, 배경에 STRUCTURE LOOK의 흰색 식당 건물과 간판을 잘못 삽입함." }, { "label": "A", "score": 4, "verdict_ko": "화면 좌측 소방차 옆에 소방관이 등장하여 '어떤 인물도 등장해선 안 된다'는 중요 지침을 위반함." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L34B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_restaurant_compound_sel.png" } ], "critique_skipped": true, "ref_mode": "플레이트+seed만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only" }, "S67sh1::bgfirst_bg": { "input_fingerprint": "2d5655119bcf4a05", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 달빛이 비치는 오르막 숲길에서 한쪽 다리를 들어 올리고 거친 숨을 내쉬며 걷는 중간 자세의 김지후(강수리영)와 정인우의 전신.\n\nLOCATION (lock): A moonlit uphill trail through dense maritime pines.\n\nTIME OF DAY (lock): night, moonlit.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height seven meters ahead, track backward on the uphill axis in a wide full-body two-shot as 김지후(강수리영) and 정인우 climb toward the lens. Hold them on opposite sides of the centerline with each caught on a different lifted leg, allowing the path between them to express both shared urgency and emotional separation.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 김지후(강수리영) in the middle-left of the frame, midground, moves toward camera-side stretch of the uphill path; 정인우 in the middle-right of the frame, midground, moves toward camera-side stretch of the uphill path.\n- KEY BACKGROUND ELEMENTS: forest path (ascending through the forest at night); used as Provides the central uphill axis for the backward tracking move; forest (dark at night); used as Encloses the pair and limits lateral escape routes.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 달빛이 비치는 오르막 숲길에서 한쪽 다리를 들어 올리고 거친 숨을 내쉬며 걷는 중간 자세의 김지후(강수리영)와 정인우의 전신.\n\nLOCATION (lock): A moonlit uphill trail through dense maritime pines.\n\nTIME OF DAY (lock): night, moonlit.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height seven meters ahead, track backward on the uphill axis in a wide full-body two-shot as 김지후(강수리영) and 정인우 climb toward the lens. Hold them on opposite sides of the centerline with each caught on a different lifted leg, allowing the path between them to express both shared urgency and emotional separation.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 김지후(강수리영) in the middle-left of the frame, midground, moves toward camera-side stretch of the uphill path; 정인우 in the middle-right of the frame, midground, moves toward camera-side stretch of the uphill path.\n- KEY BACKGROUND ELEMENTS: forest path (ascending through the forest at night); used as Provides the central uphill axis for the backward tracking move; forest (dark at night); used as Encloses the pair and limits lateral escape routes.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S67sh1__bgfirst_bg.png", "asset_id": "a850d18c-c05e-4b1d-8ad1-29db3ed28db7", "input_asset_ids": [ "5c18da0c-47ad-4484-9693-b403336cd969", "dded1895-8314-4902-9d6c-855ef980a51c" ] }, "S67sh1": { "input_fingerprint": "397feeb46d2f8177", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 달빛이 비치는 오르막 숲길에서 한쪽 다리를 들어 올리고 거친 숨을 내쉬며 걷는 중간 자세의 김지후(강수리영)와 정인우의 전신.\n\nLOCATION (lock): A moonlit uphill trail through dense maritime pines. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace, and Inwoo continues to wear the cast on his injured arm as they flee through the forest.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 달빛이 비치는 오르막 숲길에서 한쪽 다리를 들어 올리고 거친 숨을 내쉬며 걷는 중간 자세의 김지후(강수리영)와 정인우의 전신.\n\nLOCATION (lock): A moonlit uphill trail through dense maritime pines. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace, and Inwoo continues to wear the cast on his injured arm as they flee through the forest.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 달빛이 비치는 오르막 숲길에서 한쪽 다리를 들어 올리고 거친 숨을 내쉬며 걷는 중간 자세의 김지후(강수리영)와 정인우의 전신.\n\nLOCATION (lock): A moonlit uphill trail through dense maritime pines. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her bone necklace, and Inwoo continues to wear the cast on his injured arm as they flee through the forest.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S67sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S67sh1.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 '한쪽 다리를 들고 거친 숨을 내쉬는' 구체적인 동작을 두 인물 모두 입김과 포즈로 정확히 묘사하여 높은 우선순위 요건을 충족합니다." }, { "label": "B", "score": 4, "verdict_ko": "정인우의 의상은 레퍼런스에 더 가깝지만, 다리를 든 자세가 불분명하고 거친 숨을 내쉬는 묘사가 생략되어 핵심 액션 지시를 놓쳤습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 '한쪽 다리를 들고 거친 숨을 내쉬는' 구체적인 동작을 두 인물 모두 입김과 포즈로 정확히 묘사하여 높은 우선순위 요건을 충족합니다." }, { "label": "B", "score": 4, "verdict_ko": "정인우의 의상은 레퍼런스에 더 가깝지만, 다리를 든 자세가 불분명하고 거친 숨을 내쉬는 묘사가 생략되어 핵심 액션 지시를 놓쳤습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "정인우의 조끼 복장이 누락되었으나, 텍스트가 명시한 '한쪽 다리를 들어 올리고 거친 숨을 내쉬는' 구체적인 액션(입김 포함)을 정확하게 연출하여 우선순위에서 앞섭니다." }, { "label": "A", "score": 4, "verdict_ko": "캐릭터의 복장과 장비는 참조 이미지와 잘 일치하지만, 지시문이 요구한 다리를 높이 들어 올린 역동적인 자세와 거친 숨을 내쉬는 행동 묘사가 부족합니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "정인우의 조끼 복장이 누락되었으나, 텍스트가 명시한 '한쪽 다리를 들어 올리고 거친 숨을 내쉬는' 구체적인 액션(입김 포함)을 정확하게 연출하여 우선순위에서 앞섭니다." }, { "label": "B", "score": 4, "verdict_ko": "캐릭터의 복장과 장비는 참조 이미지와 잘 일치하지만, 지시문이 요구한 다리를 높이 들어 올린 역동적인 자세와 거친 숨을 내쉬는 행동 묘사가 부족합니다." } ] }, "combined": { "totals": { "A": 14, "B": 8 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 8 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 '한쪽 다리를 들고 거친 숨을 내쉬는' 구체적인 동작을 두 인물 모두 입김과 포즈로 정확히 묘사하여 높은 우선순위 요건을 충족합니다." }, { "label": "B", "score": 4, "verdict_ko": "정인우의 의상은 레퍼런스에 더 가깝지만, 다리를 든 자세가 불분명하고 거친 숨을 내쉬는 묘사가 생략되어 핵심 액션 지시를 놓쳤습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S67sh1__bgfirst_bg.png", "bg_asset_id": "a850d18c-c05e-4b1d-8ad1-29db3ed28db7", "bg_record_key": "S67sh1::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "해송_숲길", "groupbg_asset_id": "dded1895-8314-4902-9d6c-855ef980a51c" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S67sh6": { "input_fingerprint": "1904549e8488209f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 숲길을 걷는 보폭을 유지한 중간 자세로, 한 손으로 빈 목 부위를 감싸 쥔 채 당황한 표정인 김지후(강수리영)의 굳은 옆모습.\n\nLOCATION (lock): The same ascending forest path, lit by moonlight and a handheld flashlight. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same uphill pine trail, flashlight spill, moonlight, and dense undergrowth. Exclude the wider two-person walking pose and frame the woman touching her bare neck while moving.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu's neck remains bare where the bone necklace used to hang; Inwoo's injured arm remains in a cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 완벽한 옆모습보다는 반측면에 가깝지만, 이전 샷과 동일하게 빈손을 유지한 상태와 당황한 표정을 가장 충실하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "측면(옆모습) 구도는 프롬프트와 일치하나, 이전 샷에서 쥐고 있지 않던 손전등을 들고 있어 소지품 유지(Carried state) 지침을 어겼습니다." }, { "label": "C", "score": 3, "verdict_ko": "우측 가장자리에 텍스트에 명시되지 않은 타인의 팔이 등장하여 치명적인 지침 위반(Hard violation)에 해당합니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S67sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S67sh1" } }, "S67sh8": { "input_fingerprint": "10b3458faca04c45", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 숲길에 서서 낯선 사람을 보듯 김지후(강수리영)를 응시하는 정인우의 서늘한 시선.\n\nLOCATION (lock): A dark section of the uphill forest trail under sparse moonlight. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark pines, narrow uphill path, handheld flashlight light, and night atmosphere. Exclude the woman's neck-touching action and focus on the man regarding her as unfamiliar.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo continues to wear his cast while watching the necklace-less Jihu with unease.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 6, "B": 2, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지시된 남자의 차가운 시선에 온전히 집중하여 훌륭한 샷 크기를 구현했으나, 허공에 뜬 손전등이라는 물리적 오류가 감점 요인입니다." }, { "label": "B", "score": 2, "verdict_ko": "등 뒤에서 손전등을 들고 있는 정체불명의 세 번째 팔이 생성된 치명적인 신체 구조 위반(Hard Violation)입니다." }, { "label": "C", "score": 4, "verdict_ko": "남자에게 온전히 집중하라는 렌즈 구도 지시를 어기고 두 인물을 모두 담았으며, 두 사람 사이에 손전등이 공중에 떠 있는 위반이 존재합니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S67sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S67sh6" } }, "S68sh1::bgfirst_bg": { "input_fingerprint": "0dfb3ec4d94b1894", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 희뿌연 연기가 깔린 거대한 바위 제단 앞에 무릎을 꿇고 앉은 강영란(흰 머리 여자)의 무거운 뒷모습.\n\nLOCATION (lock): The ritual clearing before the massive rock altar, ringed by burning campfires and low smoke.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the establishing flow position behind 강영란(흰 머리 여자), hold a wide, eye-level composition on the central axis, with her kneeling back in the lower center and the large rock altar rising beyond her. Keep the circular fires and surrounding smoke visible at the lateral edges, making her figure feel small within the ritual space while the camera remains static before the planned reverse pan.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: large rock altar (Set before the kneeling woman for the ritual) — Its broad altar-facing side is presented toward the camera behind 강영란(흰 머리 여자); used as The primary background mass aligned directly beyond the kneeling figure; multiple campfires (Burning strongly in a circle); used as Forms a broken circular boundary around the central ritual axis; pale smoke (Surrounding the ritual area); used as Softens the outer depth while leaving the central silhouette and altar readable.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 희뿌연 연기가 깔린 거대한 바위 제단 앞에 무릎을 꿇고 앉은 강영란(흰 머리 여자)의 무거운 뒷모습.\n\nLOCATION (lock): The ritual clearing before the massive rock altar, ringed by burning campfires and low smoke.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the establishing flow position behind 강영란(흰 머리 여자), hold a wide, eye-level composition on the central axis, with her kneeling back in the lower center and the large rock altar rising beyond her. Keep the circular fires and surrounding smoke visible at the lateral edges, making her figure feel small within the ritual space while the camera remains static before the planned reverse pan.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: large rock altar (Set before the kneeling woman for the ritual) — Its broad altar-facing side is presented toward the camera behind 강영란(흰 머리 여자); used as The primary background mass aligned directly beyond the kneeling figure; multiple campfires (Burning strongly in a circle); used as Forms a broken circular boundary around the central ritual axis; pale smoke (Surrounding the ritual area); used as Softens the outer depth while leaving the central silhouette and altar readable.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S68sh1__bgfirst_bg.png", "asset_id": "0a1dced8-fead-4163-94cb-9c1b764d6b69", "input_asset_ids": [ "5bf52522-8b57-4b7e-ba88-e4ba059841a5", "68c88d61-f5d8-4e78-8c60-d1e42e155653" ] }, "S68sh1": { "input_fingerprint": "91f093d3621db313", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 희뿌연 연기가 깔린 거대한 바위 제단 앞에 무릎을 꿇고 앉은 강영란(흰 머리 여자)의 무거운 뒷모습.\n\nLOCATION (lock): The ritual clearing before the massive rock altar, ringed by burning campfires and low smoke. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran remains in her red shaman's robe before the blood-smeared rock, and the ritual fires and altar stay arranged around it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 희뿌연 연기가 깔린 거대한 바위 제단 앞에 무릎을 꿇고 앉은 강영란(흰 머리 여자)의 무거운 뒷모습.\n\nLOCATION (lock): The ritual clearing before the massive rock altar, ringed by burning campfires and low smoke. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran remains in her red shaman's robe before the blood-smeared rock, and the ritual fires and altar stay arranged around it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 희뿌연 연기가 깔린 거대한 바위 제단 앞에 무릎을 꿇고 앉은 강영란(흰 머리 여자)의 무거운 뒷모습.\n\nLOCATION (lock): The ritual clearing before the massive rock altar, ringed by burning campfires and low smoke. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran remains in her red shaman's robe before the blood-smeared rock, and the ritual fires and altar stay arranged around it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S68sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S68sh1.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION STRUCTURE PHOTOGRAPH — the confirmed photograph of this exact place and its fixed structure: it is the SINGLE authority for the location, the structure's shape, proportions, materials, colors, openings and every permanent site detail. Never copy its camera framing, time of day or lighting — the shot text is the authority for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_forest_ritual_clearing_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "지시된 '피 묻은 바위'와 짙은 안개를 잘 구현했으나, 바위 중앙의 발광 효과는 사실적인 연출을 벗어난 요소입니다." }, { "label": "A", "score": 4, "verdict_ko": "캐릭터의 뒷모습과 전체적인 구도는 지시문과 일치하지만, 필수 유지 상태인 '피 묻은 바위' 디테일이 완전히 누락되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "지시된 '피 묻은 바위'와 짙은 안개를 잘 구현했으나, 바위 중앙의 발광 효과는 사실적인 연출을 벗어난 요소입니다." }, { "label": "A", "score": 4, "verdict_ko": "캐릭터의 뒷모습과 전체적인 구도는 지시문과 일치하지만, 필수 유지 상태인 '피 묻은 바위' 디테일이 완전히 누락되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요청된 야간 배경, 모닥불, 연기, 그리고 단정히 머리를 묶은 뒷모습을 사실적으로 잘 구현했으나, 바위의 핏자국 디테일이 누락되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "바위에 피가 묻은 상태는 묘사되었으나 프롬프트에 없는 비현실적인 발광 효과가 들어갔으며, 머리를 묶은 상태라는 인물 설정과 어긋납니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요청된 야간 배경, 모닥불, 연기, 그리고 단정히 머리를 묶은 뒷모습을 사실적으로 잘 구현했으나, 바위의 핏자국 디테일이 누락되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "바위에 피가 묻은 상태는 묘사되었으나 프롬프트에 없는 비현실적인 발광 효과가 들어갔으며, 머리를 묶은 상태라는 인물 설정과 어긋납니다." } ] }, "combined": { "totals": { "A": 11, "B": 10 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 10 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "지시된 '피 묻은 바위'와 짙은 안개를 잘 구현했으나, 바위 중앙의 발광 효과는 사실적인 연출을 벗어난 요소입니다." }, { "label": "A", "score": 4, "verdict_ko": "캐릭터의 뒷모습과 전체적인 구도는 지시문과 일치하지만, 필수 유지 상태인 '피 묻은 바위' 디테일이 완전히 누락되었습니다." } ], "refs": [ { "label": "LOCATION STRUCTURE PHOTOGRAPH — the confirmed photograph of this exact place and its fixed structure: it is the SINGLE authority for the location, the structure's shape, proportions, materials, colors, openings and every permanent site detail. Never copy its camera framing, time of day or lighting — the shot text is the authority for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_forest_ritual_clearing_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S68sh1__bgfirst_bg.png", "bg_asset_id": "0a1dced8-fead-4163-94cb-9c1b764d6b69", "bg_record_key": "S68sh1::bgfirst_bg", "chain_winner": true, "authority": "seed_bg" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S68sh3": { "input_fingerprint": "dec9fb2a2784da98", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 제단 표면에 화살촉 모양의 뼈 목걸이가 막 닿은 순간의 강영란(흰 머리 여자)의 손.\n\nLOCATION (lock): At the small offering surface directly in front of the great rock in the forest ritual clearing. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rock altar, mist, firelight, and ritual arrangement. Exclude the elder's full kneeling figure and isolate her placing the bone pendant on the altar.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu's missing arrowhead-shaped bone necklace is now placed on the altar before the ritual rock and remains there under Youngran's control.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "레퍼런스의 화려한 금박 무늬 소매와 60대 여성의 손 질감을 가장 정확히 재현했으며 뼈가 제단에 닿는 순간을 잘 포착함." }, { "label": "B", "score": 5, "verdict_ko": "목걸이 끈은 표현되었으나, 캐릭터 레퍼런스의 핵심인 소매의 금박 장식이 완전히 누락되어 의상 일치도가 낮음." }, { "label": "A", "score": 3, "verdict_ko": "프롬프트에 없는 거대한 짐승의 뼈들이 제단 위에 추가되어 하드 위반(없는 사물 생성)에 해당함." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S68sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S68sh1" }, "lane_policy": "ab_select_bypass:prev" }, "S69sh3": { "input_fingerprint": "8332b6740e13eb77", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 경악하여 눈을 크게 뜬 채 그 자리에 굳어 있는 정인우의 하얗게 질린 얼굴.\n\nLOCATION (lock): A flashlight-lit stretch of dense nighttime forest away from the main trail. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark pine trail, moving flashlight illumination, dense brush, and moonlit atmosphere. Exclude the prior walking conversation and show the man frozen by his realization.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo carries the backpack and flashlight and continues to wear the cast on his injured arm. Jihu remains without the necklace, which is at Youngran's altar.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "A": 6, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "'경악하여 눈을 크게 뜬 하얗게 질린 얼굴'이라는 샷 텍스트를 클로즈업으로 완벽히 구현하였으며, 어깨에 고정된 손전등의 연속성도 정확히 유지했습니다." }, { "label": "A", "score": 6, "verdict_ko": "이전 샷의 연속성은 잘 지켰으나, 텍스트가 요구한 극심한 경악 표정과 얼굴 중심의 프레이밍이 다소 부족합니다." }, { "label": "C", "score": 3, "verdict_ko": "어깨에 고정되어 있던 손전등을 깁스한 손에 들고 있어 소지품 상태 유지(Carried state) 지침을 정면으로 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S67sh8_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S67sh8" } }, "S69sh6": { "input_fingerprint": "fc5be6f81dbb3cff", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 수풀 뒤 바닥에 바짝 엎드린 채 웅크리고 있는 정인우와 김지후(강수리영)의 굽은 등.\n\nLOCATION (lock): Low undergrowth beside a dark forest route, concealed from passing villagers. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same narrow trail, thick shrubs, pine trunks, and darkness. Exclude the upright walking figures and show the pair flattened behind the undergrowth.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu and Inwoo remain concealed flat behind the brush; Inwoo retains his backpack, flashlight, and cast, while Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "바닥에 바짝 엎드려 웅크린 자세와 굽은 등(Priority 2)을 유일하게 정확히 연출했으나, 배낭 누락과 손전등 중복(Priority 4)이 있어 감점되었습니다." }, { "label": "A", "score": 5, "verdict_ko": "수풀 뒤에 숨은 구도는 좋으나, 바짝 엎드리지 않고 쪼그려 앉은 자세(Priority 2)로 묘사되어 텍스트의 행동 지시를 충족하지 못했습니다." }, { "label": "B", "score": 3, "verdict_ko": "상체를 세우고 무릎을 꿇은 자세로 '바짝 엎드린' 지시(Priority 2)를 가장 크게 위반했으며, 소품 배치에도 오류가 많습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S69sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S69sh3" } }, "S69sh7": { "input_fingerprint": "4f54adbc1a02b906", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 등불 없이 어두운 숲길을 바삐 걸어 내려가며 한 발을 땅에서 막 뗀 mid-stride 순간의 한국인 마을 청년들의 시커먼 실루엣.\n\nLOCATION (lock): A descending trail visible between pine trunks in the unlit forest. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same black pine forest, sloping trail, brush, and moonlight. Exclude the hiding pair from the visible path and show only the unlit youths hurrying downhill.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The village youths continue moving through the forest without lights. Jihu's necklace remains at the altar, while Inwoo retains his backpack and flashlight.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "숨어있던 두 인물을 제외하라는 지시를 정확히 따랐으며, 조명 없이 산길을 내려가는 청년들의 어두운 실루엣과 걷는 동작을 잘 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "화면에서 제외해야 할 이전 장면의 인물(배낭과 손전등 착용)이 포함되어 핵심 지침을 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "프롬프트에서 명백히 제외하라고 지시한 이전 컷의 인물(팔 깁스와 손전등)이 뛰고 있어 규정을 어겼습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S69sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S69sh6" } }, "S70sh1::bgfirst_bg": { "input_fingerprint": "bfce8fa810ee58c0", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 시커먼 어둠이 입을 벌린 거대한 동굴 입구 앞에 나란히 멈춰 선 김지후(강수리영)와 정인우의 뒷모습.\n\nLOCATION (lock): The wooded ground immediately before the reinforced mouth of a large natural cave.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold the flow's static, eye-level establishing view several meters behind 김지후(강수리영) and 정인우, placing their backs side by side in the lower center and the large cave entrance directly ahead. Preserve enough surrounding space to show their hesitation, with the dark opening held as a substantial background shape without allowing it to exceed a natural environmental proportion.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: large cave entrance (Open, with a dark interior) — The camera faces into its open front, with the interior visible as darkness beyond the threshold; used as The central destination and visual void directly beyond the two figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 시커먼 어둠이 입을 벌린 거대한 동굴 입구 앞에 나란히 멈춰 선 김지후(강수리영)와 정인우의 뒷모습.\n\nLOCATION (lock): The wooded ground immediately before the reinforced mouth of a large natural cave.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold the flow's static, eye-level establishing view several meters behind 김지후(강수리영) and 정인우, placing their backs side by side in the lower center and the large cave entrance directly ahead. Preserve enough surrounding space to show their hesitation, with the dark opening held as a substantial background shape without allowing it to exceed a natural environmental proportion.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: large cave entrance (Open, with a dark interior) — The camera faces into its open front, with the interior visible as darkness beyond the threshold; used as The central destination and visual void directly beyond the two figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S70sh1__bgfirst_bg.png", "asset_id": "bcd6ac93-37d3-4038-bdd4-4cc15d51926a", "input_asset_ids": [ "19e0f73b-5078-4063-b983-1f0ad67209e5", "dad15ae4-e0d6-4247-8fc4-392b8ffcb643", "cf15bf22-d776-4059-aabd-5729e3547256" ] }, "S70sh1": { "input_fingerprint": "952b5918a4dd763e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 시커먼 어둠이 입을 벌린 거대한 동굴 입구 앞에 나란히 멈춰 선 김지후(강수리영)와 정인우의 뒷모습.\n\nLOCATION (lock): The wooded ground immediately before the reinforced mouth of a large natural cave. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her necklace, and Inwoo retains his backpack, flashlight, and arm cast at the cave entrance.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 시커먼 어둠이 입을 벌린 거대한 동굴 입구 앞에 나란히 멈춰 선 김지후(강수리영)와 정인우의 뒷모습.\n\nLOCATION (lock): The wooded ground immediately before the reinforced mouth of a large natural cave. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her necklace, and Inwoo retains his backpack, flashlight, and arm cast at the cave entrance.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 시커먼 어둠이 입을 벌린 거대한 동굴 입구 앞에 나란히 멈춰 선 김지후(강수리영)와 정인우의 뒷모습.\n\nLOCATION (lock): The wooded ground immediately before the reinforced mouth of a large natural cave. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains without her necklace, and Inwoo retains his backpack, flashlight, and arm cast at the cave entrance.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S70sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S70sh1.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L29B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_natural_cave_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "나란히 선 뒷모습과 야간 동굴 배경을 잘 구현했으며, 정인우의 백팩과 팔 깁스 상태도 지시대로 정확히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 구도와 어두운 배경 분위기는 좋으나, 명시된 정인우의 팔 깁스가 묘사되지 않아 감점되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "나란히 선 뒷모습과 야간 동굴 배경을 잘 구현했으며, 정인우의 백팩과 팔 깁스 상태도 지시대로 정확히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 구도와 어두운 배경 분위기는 좋으나, 명시된 정인우의 팔 깁스가 묘사되지 않아 감점되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 샷 구도(두 인물의 뒷모습)와 야간 배경을 잘 반영했으며, 필수 유지 항목인 정인우의 팔 깁스와 배낭을 성공적으로 묘사했습니다." }, { "label": "B", "score": 4, "verdict_ko": "뒷모습 구도와 동굴 배경은 적절하게 연출되었으나, 프롬프트에서 유지하도록 명시한 정인우의 팔 깁스가 누락되어 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 샷 구도(두 인물의 뒷모습)와 야간 배경을 잘 반영했으며, 필수 유지 항목인 정인우의 팔 깁스와 배낭을 성공적으로 묘사했습니다." }, { "label": "A", "score": 4, "verdict_ko": "뒷모습 구도와 동굴 배경은 적절하게 연출되었으나, 프롬프트에서 유지하도록 명시한 정인우의 팔 깁스가 누락되어 감점되었습니다." } ] }, "combined": { "totals": { "A": 9, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 9, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "나란히 선 뒷모습과 야간 동굴 배경을 잘 구현했으며, 정인우의 백팩과 팔 깁스 상태도 지시대로 정확히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 구도와 어두운 배경 분위기는 좋으나, 명시된 정인우의 팔 깁스가 묘사되지 않아 감점되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L29B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_natural_cave_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S70sh1__bgfirst_bg.png", "bg_asset_id": "bcd6ac93-37d3-4038-bdd4-4cc15d51926a", "bg_record_key": "S70sh1::bgfirst_bg", "chain_winner": false, "authority": "plate", "seed_attached": true }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S70sh4": { "input_fingerprint": "8c2c7d31d42202f5", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어둠 속으로 향하는 김지후(강수리영)의 뒷모습을 배경으로, 벙찐 표정으로 입을 떡 벌린 채 굳어 있는 정인우의 얼굴.\n\nLOCATION (lock): At the cave threshold, with one figure crossing from the moonlit forest into total darkness. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cave mouth, timber supports, dark forest, and moonlight. Exclude the two-person stationary pose and show the woman already entering while the man hesitates behind.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu enters without her necklace; Inwoo follows with the backpack and flashlight, his injured arm still cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "A": 6, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지후가 어둠 속으로 향하는 뒷모습을 배경으로 인우의 경악한 표정을 전면에 자연스러운 시선 처리와 함께 완벽히 담아냈습니다." }, { "label": "A", "score": 6, "verdict_ko": "요구된 구도와 소품 상태는 잘 구현했으나, 인우의 시선이 카메라 렌즈를 향하고 있어 극중 상황과 어울리지 않습니다." }, { "label": "C", "score": 3, "verdict_ko": "카메라 위치가 동굴 안에서 밖을 내다보는 완전히 반대 방향으로 설정되어 공간과 동선 지시를 심각하게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S70sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S70sh1" } }, "S71sh8::bgfirst_bg": { "input_fingerprint": "b504d303b709fc72", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 정인우의 손에 손목이 잡힌 채로 흔들림 없이 굳건한 눈빛을 발산하는 김지후(강수리영)의 단호한 얼굴.\n\nLOCATION (lock): A deep passage inside the natural cave, its bone-strewn floor revealed by a failing flashlight.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the eye-level dolly-in beside the pair, holding 김지후(강수리영)'s face close in three-quarter view while 정인우's shoulder and gripping hand enter from the near frame edge. Her restrained wrist remains visible in the lower portion as her unwavering eyes meet him, turning the physical pull and her refusal into one compact visual conflict.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) held by the wrist in the middle-center of the frame, midground; 정인우 in the lower-left of the frame, foreground, reaches for 김지후(강수리영)'s wrist.\n- KEY BACKGROUND ELEMENTS: cave interior (Dark around the pair); used as Kept indistinct behind the close confrontation to preserve the dangerous location; flashlight (Flickering as its battery weakens) — Its beam is directed toward 김지후(강수리영) from 정인우's side; used as Creates intermittent illumination across 김지후(강수리영)'s face and the restraining hand.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 정인우의 손에 손목이 잡힌 채로 흔들림 없이 굳건한 눈빛을 발산하는 김지후(강수리영)의 단호한 얼굴.\n\nLOCATION (lock): A deep passage inside the natural cave, its bone-strewn floor revealed by a failing flashlight.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the eye-level dolly-in beside the pair, holding 김지후(강수리영)'s face close in three-quarter view while 정인우's shoulder and gripping hand enter from the near frame edge. Her restrained wrist remains visible in the lower portion as her unwavering eyes meet him, turning the physical pull and her refusal into one compact visual conflict.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) held by the wrist in the middle-center of the frame, midground; 정인우 in the lower-left of the frame, foreground, reaches for 김지후(강수리영)'s wrist.\n- KEY BACKGROUND ELEMENTS: cave interior (Dark around the pair); used as Kept indistinct behind the close confrontation to preserve the dangerous location; flashlight (Flickering as its battery weakens) — Its beam is directed toward 김지후(강수리영) from 정인우's side; used as Creates intermittent illumination across 김지후(강수리영)'s face and the restraining hand.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S71sh8__bgfirst_bg.png", "asset_id": "3e132205-6224-4f43-9be8-83d3d6ab8985", "input_asset_ids": [ "7bd8938e-fe31-4bee-9cfa-0ccf5e57c9fd", "d8a051b0-2dff-4d88-a8b2-b532022bab38" ] }, "S71sh8": { "input_fingerprint": "18a53d5d17f3cd1a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정인우의 손에 손목이 잡힌 채로 흔들림 없이 굳건한 눈빛을 발산하는 김지후(강수리영)의 단호한 얼굴.\n\nLOCATION (lock): A deep passage inside the natural cave, its bone-strewn floor revealed by a failing flashlight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo keeps the failing flashlight and backpack while his injured arm remains cast; Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정인우의 손에 손목이 잡힌 채로 흔들림 없이 굳건한 눈빛을 발산하는 김지후(강수리영)의 단호한 얼굴.\n\nLOCATION (lock): A deep passage inside the natural cave, its bone-strewn floor revealed by a failing flashlight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo keeps the failing flashlight and backpack while his injured arm remains cast; Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 정인우의 손에 손목이 잡힌 채로 흔들림 없이 굳건한 눈빛을 발산하는 김지후(강수리영)의 단호한 얼굴.\n\nLOCATION (lock): A deep passage inside the natural cave, its bone-strewn floor revealed by a failing flashlight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo keeps the failing flashlight and backpack while his injured arm remains cast; Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S71sh8__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S71sh8.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B07.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 구도와 인물의 단호한 표정, 신체 해부학적 구조를 정확히 구현했으나, 인우의 깁스가 누락되고 지후의 목걸이가 남아있어 소품 지시에서 감점되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "깁스와 모자 등 요구된 소품 상태는 잘 반영했으나, 정인우가 지후를 잡은 오른쪽 팔이 구조상 왼손으로 그려져 치명적인 해부학적 위반(Hard Violation)에 해당합니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 구도와 인물의 단호한 표정, 신체 해부학적 구조를 정확히 구현했으나, 인우의 깁스가 누락되고 지후의 목걸이가 남아있어 소품 지시에서 감점되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "깁스와 모자 등 요구된 소품 상태는 잘 반영했으나, 정인우가 지후를 잡은 오른쪽 팔이 구조상 왼손으로 그려져 치명적인 해부학적 위반(Hard Violation)에 해당합니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지후의 단호한 얼굴과 손목을 잡은 앵글을 잘 연출했으며, 인우의 깁스와 지후의 목걸이 미착용 등 디테일한 상태 지시를 정확히 따랐습니다." }, { "label": "A", "score": 4, "verdict_ko": "지후의 표정 연출은 좋으나, 인우의 깁스가 누락되었고 지후가 목걸이를 착용하고 있어 상태 유지(Carried State) 지침을 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지후의 단호한 얼굴과 손목을 잡은 앵글을 잘 연출했으며, 인우의 깁스와 지후의 목걸이 미착용 등 디테일한 상태 지시를 정확히 따랐습니다." }, { "label": "B", "score": 4, "verdict_ko": "지후의 표정 연출은 좋으나, 인우의 깁스가 누락되었고 지후가 목걸이를 착용하고 있어 상태 유지(Carried State) 지침을 위반했습니다." } ] }, "combined": { "totals": { "A": 11, "B": 11 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 11 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 구도와 인물의 단호한 표정, 신체 해부학적 구조를 정확히 구현했으나, 인우의 깁스가 누락되고 지후의 목걸이가 남아있어 소품 지시에서 감점되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "깁스와 모자 등 요구된 소품 상태는 잘 반영했으나, 정인우가 지후를 잡은 오른쪽 팔이 구조상 왼손으로 그려져 치명적인 해부학적 위반(Hard Violation)에 해당합니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B07.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S71sh8__bgfirst_bg.png", "bg_asset_id": "3e132205-6224-4f43-9be8-83d3d6ab8985", "bg_record_key": "S71sh8::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S71sh13": { "input_fingerprint": "c8f605032d20526d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 동굴 구석, 목과 팔다리가 굵은 쇠사슬에 묶인 채 바닥에 웅크려 앉은 처참한 몰골의 강민숙(홍은미)의 전신.\n\nLOCATION (lock): A bloodstained holding recess deep inside the cave, illuminated only by a handheld flashlight. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cave wall, floor texture, low light, and heavy chain fixtures. Exclude the standing man from the earlier shot and reveal the chained woman crouched in the corner.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is crouched on the cave floor in a tight seated curl, with her battered torso folded forward, her head lowered, and her neck and limbs restrained by heavy chains against the corner.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains chained in the cave corner in the same weakened condition. Jihu lacks her necklace, while Inwoo retains his backpack, flashlight, and cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "샷 텍스트에 없는 인물을 프레임에서 완벽히 배제하고 조명 효과만 활용하여 쇠사슬에 묶인 여성의 전신을 가장 정확하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "이전 샷의 남성을 제외하라는 지시를 어기고 남성의 팔과 몸통 일부가 화면에 크게 노출되는 치명적인 위반을 범했습니다." }, { "label": "C", "score": 4, "verdict_ko": "텍스트에 명시되지 않은 인물의 손과 손전등이 화면에 직접 등장하여 '지정된 인물 외 추가 금지' 규칙을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S71sh8_sel.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S71sh8" } }, "S71sh16": { "input_fingerprint": "f4cb167afd15aa10", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 고통과 원망이 섞인 표정으로 얼굴을 일그러뜨린 채 입을 크게 벌리고 오열하는 강민숙(홍은미)의 붉어진 얼굴.\n\nLOCATION (lock): The chained corner of the cave chamber amid damp rock, bones, and blood traces. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dim cave, rough stone, chains, and damp atmosphere. Exclude the man's earlier position and isolate the chained woman's anguished reaction.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is crouched on the cave floor in a tight seated curl, with her battered torso folded forward, her head lowered, and her neck and limbs restrained by heavy chains against the corner.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains restrained by the cave's heavy chain while crying; Jihu remains necklace-less, and Inwoo keeps his backpack and cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 4, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지문이 요구한 '붉어진 얼굴' 클로즈업 샷 크기를 정확히 구현하여, 더 넓게 렌더링한 다른 후보들을 압도합니다." }, { "label": "A", "score": 4, "verdict_ko": "얼굴 묘사는 지문에 부합하나, 지문이 요구한 얼굴 위주의 샷 크기를 무시하고 전신 롱샷을 렌더링하여 감점이 큽니다." }, { "label": "C", "score": 3, "verdict_ko": "샷 크기 지시를 위반하여 전신을 보여주었으며, 표정의 강도도 지문(크게 벌리고 오열)에 비해 약합니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S71sh13_sel.png" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S71sh13" }, "locked_char_refs_excluded": [ "강민숙(홍은미)(C09)" ] }, "S72sh1": { "input_fingerprint": "459b857120ebbe70", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 뼈대만 남은 채 시뻘건 불씨와 거뭇한 연기를 뿜어내고 있는 전소된 창고의 전경.\n\nLOCATION (lock): The restaurant’s rear yard around the charred remains of the burned storage shed. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The restaurant's rear shed is now completely burned down, with only its frame, embers, and dark smoke remaining.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 5, "A": 4 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "'뼈대만 남은 채'라는 지문을 가장 정확히 반영하여 창고가 뼈대만 남은 상태를 훌륭하게 구현했습니다." }, { "label": "C", "score": 5, "verdict_ko": "뼈대가 드러났으나 벽면 패널이 여전히 많이 남아 있어 전소된 상태의 묘사가 다소 부족합니다." }, { "label": "A", "score": 4, "verdict_ko": "창고의 외벽이 대부분 그대로 남아 있어 '뼈대만 남은'이라는 핵심 지문을 충족하지 못했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S66sh1_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S66sh1" }, "lane_policy": "ab_select_bypass:bg_only:share_plan_prev_bgonly" }, "S72sh2": { "input_fingerprint": "ad9fdaa317544507", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 불타버린 창고를 바라보며 넋이 나간 듯 허탈한 표정을 짓고 있는 황봉업(황 씨·민숙 집의 거구 남자)의 얼굴.\n\nLOCATION (lock): Beside the smoldering shed ruins in the restaurant’s rear yard. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same charred shed frame, embers, black smoke, and night lighting. Exclude the empty establishing composition and place the large man staring at the ruin.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The destroyed shed remains a smoking, charred ruin behind the restaurant.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 '얼굴' 중심의 클로즈업 샷 크기를 가장 잘 구현했고 허탈한 표정이 돋보이나, 창고를 등진 시선 방향이 다소 아쉽습니다." }, { "label": "A", "score": 5, "verdict_ko": "인물과 배경의 세부 묘사는 안정적이나, 프레임이 상반신 전체로 넓어져 '얼굴' 중심의 샷 지시를 벗어났습니다." }, { "label": "B", "score": 3, "verdict_ko": "표정과 샷 크기는 적절하나, 배경 우측 벽면에 불타는 창고 지붕이 중복으로 생성되는 치명적 오류(Hard Violation)가 발생했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S72sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S72sh1" } }, "S73sh1::bgfirst_bg": { "input_fingerprint": "a9dbfc0dc1043f0a", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 타오르는 커다란 모닥불 주위 거대한 바위 옆에서 한 발을 내딛고 몸을 튼 채, 한 손으로 무령을 허공에 세차게 치켜든 강영란(흰 머리 여자)의 상체 자세.\n\nLOCATION (lock): The firelit ritual clearing around the massive forest rock altar at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track laterally at torso height along the outer edge of Youngran's path beside the large rock, tightening to a medium upper-body frame without changing the established movement direction. Catch her weight transferring into the turn, with the raised ritual bells held in the upper third, the bonfire offset behind her, and part of the rock anchoring the opposite edge.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground, moves toward path around the large rock; large rock in the middle-left of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 거대한 바위 (Beside Youngran's path); used as Side anchor that reveals the curved ritual path and supports the tracking direction; 커다란 모닥불 (Burning strongly); used as Background ritual focus kept offset from Youngran's raised arm; 북과 징 (Being played by the musicians) — Their played sides are angled toward the ritual musicians in the background; used as Peripheral context for the ongoing rite.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is an open ritual clearing enclosed on every side by forest. A towering giant rock occupies the rear of the principal ritual forecourt and faces into the open ground. The central altar stands immediately in front of the rock on the clearing side, creating the main spatial axis: outer clearing, altar, then giant rock. Open ground extends laterally around this forecourt and outward to the surrounding forest edge, allowing movement around the rock and assembly across the wider clearing. A traversable forest-side approach enters the outer clearing through the enclosing woodland, although the source does not establish its compass direction. No permanent fire structures are specified; bonfires, smoke, blood traces, restraints, offerings, equipment, and participant formations remain moment-specific additions rather than site markers.\n- clearing: forest-enclosed ritual clearing\n- rock: towering giant ritual rock\n- altar: central altar before the giant rock\n- boundary: surrounding forest edge\n- approach: traversable forest-side approach\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 타오르는 커다란 모닥불 주위 거대한 바위 옆에서 한 발을 내딛고 몸을 튼 채, 한 손으로 무령을 허공에 세차게 치켜든 강영란(흰 머리 여자)의 상체 자세.\n\nLOCATION (lock): The firelit ritual clearing around the massive forest rock altar at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track laterally at torso height along the outer edge of Youngran's path beside the large rock, tightening to a medium upper-body frame without changing the established movement direction. Catch her weight transferring into the turn, with the raised ritual bells held in the upper third, the bonfire offset behind her, and part of the rock anchoring the opposite edge.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground, moves toward path around the large rock; large rock in the middle-left of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 거대한 바위 (Beside Youngran's path); used as Side anchor that reveals the curved ritual path and supports the tracking direction; 커다란 모닥불 (Burning strongly); used as Background ritual focus kept offset from Youngran's raised arm; 북과 징 (Being played by the musicians) — Their played sides are angled toward the ritual musicians in the background; used as Peripheral context for the ongoing rite.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is an open ritual clearing enclosed on every side by forest. A towering giant rock occupies the rear of the principal ritual forecourt and faces into the open ground. The central altar stands immediately in front of the rock on the clearing side, creating the main spatial axis: outer clearing, altar, then giant rock. Open ground extends laterally around this forecourt and outward to the surrounding forest edge, allowing movement around the rock and assembly across the wider clearing. A traversable forest-side approach enters the outer clearing through the enclosing woodland, although the source does not establish its compass direction. No permanent fire structures are specified; bonfires, smoke, blood traces, restraints, offerings, equipment, and participant formations remain moment-specific additions rather than site markers.\n- clearing: forest-enclosed ritual clearing\n- rock: towering giant ritual rock\n- altar: central altar before the giant rock\n- boundary: surrounding forest edge\n- approach: traversable forest-side approach\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S73sh1__bgfirst_bg.png", "asset_id": "fc6e673c-19b5-461c-8fad-d43dc8730e6b", "input_asset_ids": [ "30d1e2f3-47b3-496f-b9be-3ca9ae15d8a4" ] }, "S73sh1": { "input_fingerprint": "90cbf71f8bbc8c6e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 타오르는 커다란 모닥불 주위 거대한 바위 옆에서 한 발을 내딛고 몸을 튼 채, 한 손으로 무령을 허공에 세차게 치켜든 강영란(흰 머리 여자)의 상체 자세.\n\nLOCATION (lock): The firelit ritual clearing around the massive forest rock altar at night. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same giant rock, circular fires, smoky clearing, and nocturnal ritual light. Exclude the kneeling prayer pose and show the elder circling with the ritual bell raised.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe, and the fires, instruments, and blood-marked ritual rock remain set at the ceremonial clearing. Jihu's necklace remains on the altar.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 타오르는 커다란 모닥불 주위 거대한 바위 옆에서 한 발을 내딛고 몸을 튼 채, 한 손으로 무령을 허공에 세차게 치켜든 강영란(흰 머리 여자)의 상체 자세.\n\nLOCATION (lock): The firelit ritual clearing around the massive forest rock altar at night. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe, and the fires, instruments, and blood-marked ritual rock remain set at the ceremonial clearing. Jihu's necklace remains on the altar.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 타오르는 커다란 모닥불 주위 거대한 바위 옆에서 한 발을 내딛고 몸을 튼 채, 한 손으로 무령을 허공에 세차게 치켜든 강영란(흰 머리 여자)의 상체 자세.\n\nLOCATION (lock): The firelit ritual clearing around the massive forest rock altar at night. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe, and the fires, instruments, and blood-marked ritual rock remain set at the ceremonial clearing. Jihu's necklace remains on the altar.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 타오르는 커다란 모닥불 주위 거대한 바위 옆에서 한 발을 내딛고 몸을 튼 채, 한 손으로 무령을 허공에 세차게 치켜든 강영란(흰 머리 여자)의 상체 자세.\n\nLOCATION (lock): The firelit ritual clearing around the massive forest rock altar at night. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe, and the fires, instruments, and blood-marked ritual rock remain set at the ceremonial clearing. Jihu's necklace remains on the altar.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S73sh1__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S73sh1__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S73sh1__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "A": 6, "B": 4, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "상체 샷 지시를 놓쳐 전신이 나왔으나, 레퍼런스의 붉은 모자와 묶은 머리 설정을 유일하게 모두 충족했습니다." }, { "label": "B", "score": 4, "verdict_ko": "상체 샷 지시를 어겼으며, 의상 레퍼런스의 핵심인 붉은 모자가 완전히 누락되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "상체 샷 지시를 어겼으며, '단정하게 묶은 머리'라는 캐릭터 설정과 정반대로 머리를 길게 늘어뜨렸습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S73sh1__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S73sh1__bgfirst_bg.png", "bg_asset_id": "fc6e673c-19b5-461c-8fad-d43dc8730e6b", "bg_record_key": "S73sh1::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S68sh3" } }, "S73sh4": { "input_fingerprint": "6e60f9ca2a3d18b2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 심각하게 굳은 표정으로 서로의 눈을 똑바로 응시하는 고산들과 강영란(흰 머리 여자)의 교차된 시선.\n\nLOCATION (lock): At the edge of the active ritual ground, between the musicians and the rock altar. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same firelit rock, smoke, ritual musicians, and dark forest perimeter. Exclude the elder's dancing pose and frame her exchanging a grave look with the man.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran remains in her red shaman's robe, with Jihu's bone necklace still held at the ritual altar.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 5, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "의상(여성의 붉은 모자 포함)과 소지품(뼈 목걸이) 지시를 가장 충실히 따랐고, 두 인물의 심각한 시선 교환을 정확히 연출했습니다." }, { "label": "A", "score": 5, "verdict_ko": "남성의 의상과 배경은 일치하나, 여성의 모자가 누락되었고 뼈 목걸이 대신 방울을 들고 있어 소지품 지시를 위반했습니다." }, { "label": "C", "score": 4, "verdict_ko": "여성이 뼈 목걸이를 들고 있으나 모자가 누락되었고, 남성이 참고 이미지와 완전히 다른 검은색 재킷을 입고 있어 감점되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S73sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S73sh1" } }, "S74sh2::bgfirst_bg": { "input_fingerprint": "52b116e0b9d69108", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 걸음을 멈춘 채 고개를 뒤로 돌려 분노로 눈을 부릅뜬 강민숙(홍은미)의 매서운 얼굴.\n\nLOCATION (lock): A dark escape route through dense island forest away from the cave.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the end of the curved track, settle at Minsook's eye level on her front three-quarter side and move slightly closer until her fierce face becomes the primary close subject. Her stopped body remains twisted back toward Suriyoung, with a sliver of the unlit forest path behind her shoulder preserving the direction from which the group came.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 숲길 (Dark, with no light visible); used as Narrow background corridor preserving the group's direction of travel; 숲 (Dark); used as Dense background boundary around Minsook's turned silhouette.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 걸음을 멈춘 채 고개를 뒤로 돌려 분노로 눈을 부릅뜬 강민숙(홍은미)의 매서운 얼굴.\n\nLOCATION (lock): A dark escape route through dense island forest away from the cave.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the end of the curved track, settle at Minsook's eye level on her front three-quarter side and move slightly closer until her fierce face becomes the primary close subject. Her stopped body remains twisted back toward Suriyoung, with a sliver of the unlit forest path behind her shoulder preserving the direction from which the group came.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 숲길 (Dark, with no light visible); used as Narrow background corridor preserving the group's direction of travel; 숲 (Dark); used as Dense background boundary around Minsook's turned silhouette.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S74sh2__bgfirst_bg.png", "asset_id": "aa837d06-718e-49ed-94bc-66c888af63a7", "input_asset_ids": [ "54716564-8572-4460-b16e-b3e9929532cb", "dded1895-8314-4902-9d6c-855ef980a51c" ] }, "S74sh2": { "input_fingerprint": "03e98a5593577807", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 걸음을 멈춘 채 고개를 뒤로 돌려 분노로 눈을 부릅뜬 강민숙(홍은미)의 매서운 얼굴.\n\nLOCATION (lock): A dark escape route through dense island forest away from the cave. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook is no longer fixed to the cave spot and leads the escape through the forest; Jihu remains without her necklace, while Inwoo retains his backpack and arm cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 걸음을 멈춘 채 고개를 뒤로 돌려 분노로 눈을 부릅뜬 강민숙(홍은미)의 매서운 얼굴.\n\nLOCATION (lock): A dark escape route through dense island forest away from the cave. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook is no longer fixed to the cave spot and leads the escape through the forest; Jihu remains without her necklace, while Inwoo retains his backpack and arm cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 걸음을 멈춘 채 고개를 뒤로 돌려 분노로 눈을 부릅뜬 강민숙(홍은미)의 매서운 얼굴.\n\nLOCATION (lock): A dark escape route through dense island forest away from the cave. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook is no longer fixed to the cave spot and leads the escape through the forest; Jihu remains without her necklace, while Inwoo retains his backpack and arm cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S74sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S74sh2.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지문에 명시된 '고개를 뒤로 돌려 매서운 얼굴'이라는 핵심 포즈와 표정을 정확히 구현했으나, 샷 텍스트에 없는 인물들이 배경에 등장한 점은 감점 요소입니다." }, { "label": "B", "score": 4, "verdict_ko": "'고개를 뒤로 돌려'라는 묘사와 달리 정면을 향해 서 있으며, 이 후보 역시 배경에 불필요한 인물들이 포함되어 지시를 어겼습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지문에 명시된 '고개를 뒤로 돌려 매서운 얼굴'이라는 핵심 포즈와 표정을 정확히 구현했으나, 샷 텍스트에 없는 인물들이 배경에 등장한 점은 감점 요소입니다." }, { "label": "B", "score": 4, "verdict_ko": "'고개를 뒤로 돌려'라는 묘사와 달리 정면을 향해 서 있으며, 이 후보 역시 배경에 불필요한 인물들이 포함되어 지시를 어겼습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "지문이 요구한 '고개를 뒤로 돌려' 돌아보는 핵심 동작을 올바르게 연출했으나, 분노한 표정이 약하고 지문에 없는 인물들이 배경에 포함되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "분노로 부릅뜬 표정은 잘 표현되었으나, '고개를 뒤로 돌려' 돌아보는 동작을 누락한 채 정면을 향해 있으며 역시 불필요한 인물들이 등장합니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지문이 요구한 '고개를 뒤로 돌려' 돌아보는 핵심 동작을 올바르게 연출했으나, 분노한 표정이 약하고 지문에 없는 인물들이 배경에 포함되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "분노로 부릅뜬 표정은 잘 표현되었으나, '고개를 뒤로 돌려' 돌아보는 동작을 누락한 채 정면을 향해 있으며 역시 불필요한 인물들이 등장합니다." } ] }, "combined": { "totals": { "A": 14, "B": 8 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 8 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지문에 명시된 '고개를 뒤로 돌려 매서운 얼굴'이라는 핵심 포즈와 표정을 정확히 구현했으나, 샷 텍스트에 없는 인물들이 배경에 등장한 점은 감점 요소입니다." }, { "label": "B", "score": 4, "verdict_ko": "'고개를 뒤로 돌려'라는 묘사와 달리 정면을 향해 서 있으며, 이 후보 역시 배경에 불필요한 인물들이 포함되어 지시를 어겼습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S74sh2__bgfirst_bg.png", "bg_asset_id": "aa837d06-718e-49ed-94bc-66c888af63a7", "bg_record_key": "S74sh2::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "해송_숲길", "groupbg_asset_id": "dded1895-8314-4902-9d6c-855ef980a51c" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S74sh8": { "input_fingerprint": "1d19cf8b565d0d74", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 숲길에 선 채 먼 곳에서 들려온 휘파람 소리에 놀라 고개를 반쯤 뒤로 돌린 김지후(강수리영)의 굳은 옆모습.\n\nLOCATION (lock): The same unlit forest route as the fugitives hurry downhill. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same shadowy pines, narrow trail, moonlight, and undergrowth. Exclude the earlier conversation pose and show the daughter turning toward the distant whistle.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The three escapees continue through the forest together; Jihu remains without her necklace, and Inwoo keeps his backpack and cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 4, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "이전 샷에서 확립된 지후의 복장(회녹색 셔츠)과 헤어스타일을 정확히 유지하며, 뒤돌아보는 굳은 옆모습을 지시문대로 잘 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 얼굴은 일치하나, 이전 샷에서 설정된 셔츠와 묶은 머리 대신 다른 의상과 풀어진 머리로 렌더링되어 연속성이 어긋납니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트가 단독으로 요구한 김지후(딸)가 아닌, 이전 샷의 찢어진 옷을 입은 다른 인물을 등장시켜 심각한 오류를 범했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S74sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S74sh2" } }, "S75sh2": { "input_fingerprint": "dbacdf51d34d823f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 횃불 불빛 아래, 텅 빈 바닥에 덩그러니 놓인 채 거칠게 끊어져 있는 굵은 쇠사슬의 클로즈업.\n\nLOCATION (lock): The abandoned holding spot inside the cave, now lit by villagers’ torches. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The heavy chain that restrained Minsook remains broken and abandoned at her now-empty place in the cave.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 5 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "'거칠게 끊어진 굵은 쇠사슬'이라는 묘사에 맞춰 사슬 고리가 단선된 모습을 클로즈업으로 가장 정확하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "사슬 자체가 아닌 족쇄가 끊어진 모습으로 묘사되어 '끊어진 쇠사슬'이라는 지문의 세부 내용과 다소 차이가 있습니다." }, { "label": "B", "score": 5, "verdict_ko": "A와 마찬가지로 사슬 고리가 아닌 족쇄 파손에 초점을 맞춰 텍스트의 묘사를 온전히 반영하지 못했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B02.png" } ], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S76sh3::bgfirst_bg": { "input_fingerprint": "2fe4584574cf6a40", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 어둠 속 달빛이 비치는 염소 도축장을 향해 방향을 틀고 급히 한 발을 허공에 뻗은 강민숙(홍은미)의 역동적인 뒷모습.\n\nLOCATION (lock): A forest path immediately beside the slaughterhouse and nearby goat pens at night.\n\nTIME OF DAY (lock): night, moonlit.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track at waist height behind and slightly beside Minsook in a wide, medium-full-body view as she pivots off the road and launches toward the slaughterhouse. Freeze the key instant with one foot airborne, her back driving diagonally from the lower foreground toward the entrance in the background, while the road remains visible as a scale reference.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강민숙(홍은미) in the middle-center of the frame, midground, moves toward slaughterhouse entrance; slaughterhouse entrance in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 염소 도축장 (Visible in the darkness under moonlight) — The entrance side faces Minsook's approach and remains visible ahead of her; used as Destination of Minsook's diagonal movement; 도축장 옆 숲길 (Occupied by the stopped group before Minsook runs); used as Ground plane preserving the sudden change from forward travel to a sideward dash; 축사 (Goats are crying loudly inside) — Its side adjoining the slaughterhouse is visible from the road; used as Adjacent source area for the goats' cries.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is a slaughterhouse compound contained within woodland. A path runs through the wooded surroundings and passes directly beside the compound, creating the exterior approach used to reach the slaughterhouse. The main slaughterhouse building occupies the compound edge next to this path. A path-side entrance connects that exterior approach to the building interior. A livestock shed is also part of the same compound near the main building, although its exact side relative to the building is not fixed by the scene. The practical camera geography therefore consists of a receding woodland path beside a building frontage, with the entrance as the principal connection between them.\n- path: woodland path beside the slaughterhouse\n- woodland: wooded surroundings around the slaughterhouse path\n- building: goat slaughterhouse building\n- entrance: path-side entrance into the slaughterhouse\n- livestock shed: livestock shed within the slaughterhouse compound\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 어둠 속 달빛이 비치는 염소 도축장을 향해 방향을 틀고 급히 한 발을 허공에 뻗은 강민숙(홍은미)의 역동적인 뒷모습.\n\nLOCATION (lock): A forest path immediately beside the slaughterhouse and nearby goat pens at night.\n\nTIME OF DAY (lock): night, moonlit.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track at waist height behind and slightly beside Minsook in a wide, medium-full-body view as she pivots off the road and launches toward the slaughterhouse. Freeze the key instant with one foot airborne, her back driving diagonally from the lower foreground toward the entrance in the background, while the road remains visible as a scale reference.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강민숙(홍은미) in the middle-center of the frame, midground, moves toward slaughterhouse entrance; slaughterhouse entrance in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 염소 도축장 (Visible in the darkness under moonlight) — The entrance side faces Minsook's approach and remains visible ahead of her; used as Destination of Minsook's diagonal movement; 도축장 옆 숲길 (Occupied by the stopped group before Minsook runs); used as Ground plane preserving the sudden change from forward travel to a sideward dash; 축사 (Goats are crying loudly inside) — Its side adjoining the slaughterhouse is visible from the road; used as Adjacent source area for the goats' cries.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is a slaughterhouse compound contained within woodland. A path runs through the wooded surroundings and passes directly beside the compound, creating the exterior approach used to reach the slaughterhouse. The main slaughterhouse building occupies the compound edge next to this path. A path-side entrance connects that exterior approach to the building interior. A livestock shed is also part of the same compound near the main building, although its exact side relative to the building is not fixed by the scene. The practical camera geography therefore consists of a receding woodland path beside a building frontage, with the entrance as the principal connection between them.\n- path: woodland path beside the slaughterhouse\n- woodland: wooded surroundings around the slaughterhouse path\n- building: goat slaughterhouse building\n- entrance: path-side entrance into the slaughterhouse\n- livestock shed: livestock shed within the slaughterhouse compound\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S76sh3__bgfirst_bg.png", "asset_id": "f16e81f3-ce06-4bd0-8b1e-5f56495fa7fc", "input_asset_ids": [ "b2511aab-d26d-4e8f-b87a-5fe066338bd0" ] }, "S76sh3": { "input_fingerprint": "a318541a2c19b817", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 어둠 속 달빛이 비치는 염소 도축장을 향해 방향을 틀고 급히 한 발을 허공에 뻗은 강민숙(홍은미)의 역동적인 뒷모습.\n\nLOCATION (lock): A forest path immediately beside the slaughterhouse and nearby goat pens at night. The shot takes place here — the attached STORYBOARD SKETCH fixes the staging, camera and figure placement of this exact place. No location photograph is attached — build the location itself strictly from the location text above and the shot text, inventing nothing beyond them.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook, Jihu, and Inwoo remain together during the escape; Jihu lacks her necklace, while Inwoo retains his backpack and cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 어둠 속 달빛이 비치는 염소 도축장을 향해 방향을 틀고 급히 한 발을 허공에 뻗은 강민숙(홍은미)의 역동적인 뒷모습.\n\nLOCATION (lock): A forest path immediately beside the slaughterhouse and nearby goat pens at night. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook, Jihu, and Inwoo remain together during the escape; Jihu lacks her necklace, while Inwoo retains his backpack and cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 어둠 속 달빛이 비치는 염소 도축장을 향해 방향을 틀고 급히 한 발을 허공에 뻗은 강민숙(홍은미)의 역동적인 뒷모습.\n\nLOCATION (lock): A forest path immediately beside the slaughterhouse and nearby goat pens at night. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook, Jihu, and Inwoo remain together during the escape; Jihu lacks her necklace, while Inwoo retains his backpack and cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 어둠 속 달빛이 비치는 염소 도축장을 향해 방향을 틀고 급히 한 발을 허공에 뻗은 강민숙(홍은미)의 역동적인 뒷모습.\n\nLOCATION (lock): A forest path immediately beside the slaughterhouse and nearby goat pens at night. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook, Jihu, and Inwoo remain together during the escape; Jihu lacks her necklace, while Inwoo retains his backpack and cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S76sh3__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S76sh3__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S76sh3__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "B": 8, "C": 6, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "스토리보드와 동일한 구도에서 캐릭터의 지정된 의상과 역동적인 뒷모습을 가장 정확하게 구현했습니다." }, { "label": "C", "score": 6, "verdict_ko": "의상과 배경은 일치하지만, 고개를 돌려 측면 얼굴이 드러나 완벽한 뒷모습을 요구한 지시와 스토리보드 구도에서 다소 벗어났습니다." }, { "label": "A", "score": 3, "verdict_ko": "배경과 뒷모습 구도는 따랐으나, 캐릭터가 레퍼런스와 완전히 다른 현대식 트레이닝복을 입고 있어 감점되었습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S76sh3__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S76sh3__bgfirst_bg.png", "bg_asset_id": "f16e81f3-ce06-4bd0-8b1e-5f56495fa7fc", "bg_record_key": "S76sh3::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+엔티티", "share_plan": { "ref_plan": "background" } }, "S76sh6": { "input_fingerprint": "b178c37f8742ec4f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, moonlit.\n\nSHOT TEXT (authoritative, Korean): 어둠 속 텅 빈 숲길에 덩그러니 선 채 어이없다는 듯 입을 반쯤 벌린 김지후(강수리영)의 멍한 표정.\n\nLOCATION (lock): The dark path outside the slaughterhouse entrance under moonlight. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same moonlit slaughterhouse exterior, adjoining forest trail, and dark goat-pen surroundings. Exclude the mother rushing into the building and leave the daughter standing in bafflement.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook has entered the slaughterhouse while Jihu and Inwoo remain just outside; Jihu is necklace-less, and Inwoo still carries the backpack and wears the cast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 5, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "'멍한 표정'에 집중해야 하는 지시문에 맞게 클로즈업 프레이밍을 완벽히 선택했으며, 벌어진 입의 디테일도 잘 살렸습니다." }, { "label": "A", "score": 5, "verdict_ko": "장소와 의상의 연속성은 좋으나, '표정'을 강조하기에는 앵글이 너무 넓고 불필요하게 뒤돌아보는 포즈를 취했습니다." }, { "label": "C", "score": 4, "verdict_ko": "전신을 다 보여주는 풀샷으로 구도를 넓히는 바람에 '표정'을 묘사하라는 샷의 가장 중요한 의도에서 크게 벗어났습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S76sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S76sh3" } }, "S77sh1::bgfirst_bg": { "input_fingerprint": "326b1f88c53578a1", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 도축장 안, 양동이에 가득 담긴 검붉은 염소 피를 자신의 머리 위로 쏟아붓는 도중인 강민숙(홍은미)의 젖은 상체.\n\nLOCATION (lock): Inside the dark slaughterhouse beside buckets of collected goat blood.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold static at chest height directly in front of Minsook in a medium upper-body composition, keeping the raised bucket inside the upper edge as goat blood falls over her head and torso. Place her wet body centrally but allow strips of the dark slaughterhouse interior on both sides; the bucket remains a secondary element above her rather than occupying excessive frame area.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 양동이 (Being emptied of goat blood) — Its open side tips downward over Minsook's head while the outer side remains visible to camera; used as Action prop retained in the upper portion of the frame; 검붉은 염소 피 (Pouring from the bucket and covering Minsook); used as Falling action layer across Minsook's head and upper body; 도축장 내부 (Dark); used as Dark negative space framing the central action.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 도축장 안, 양동이에 가득 담긴 검붉은 염소 피를 자신의 머리 위로 쏟아붓는 도중인 강민숙(홍은미)의 젖은 상체.\n\nLOCATION (lock): Inside the dark slaughterhouse beside buckets of collected goat blood.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold static at chest height directly in front of Minsook in a medium upper-body composition, keeping the raised bucket inside the upper edge as goat blood falls over her head and torso. Place her wet body centrally but allow strips of the dark slaughterhouse interior on both sides; the bucket remains a secondary element above her rather than occupying excessive frame area.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 양동이 (Being emptied of goat blood) — Its open side tips downward over Minsook's head while the outer side remains visible to camera; used as Action prop retained in the upper portion of the frame; 검붉은 염소 피 (Pouring from the bucket and covering Minsook); used as Falling action layer across Minsook's head and upper body; 도축장 내부 (Dark); used as Dark negative space framing the central action.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S77sh1__bgfirst_bg.png", "asset_id": "40b4d9b1-cd1c-4960-b963-c912d4fe5fde", "input_asset_ids": [ "48914646-a738-42d0-8634-56bf6f675b0d", "0cf5113e-e387-4d64-9d8b-6afe9e7e8352" ] }, "S77sh1": { "input_fingerprint": "5d9c53c12680bb8d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 도축장 안, 양동이에 가득 담긴 검붉은 염소 피를 자신의 머리 위로 쏟아붓는 도중인 강민숙(홍은미)의 젖은 상체.\n\nLOCATION (lock): Inside the dark slaughterhouse beside buckets of collected goat blood. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook becomes completely coated in fresh goat blood, which must remain on her body and clothing as scent camouflage.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 도축장 안, 양동이에 가득 담긴 검붉은 염소 피를 자신의 머리 위로 쏟아붓는 도중인 강민숙(홍은미)의 젖은 상체.\n\nLOCATION (lock): Inside the dark slaughterhouse beside buckets of collected goat blood. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook becomes completely coated in fresh goat blood, which must remain on her body and clothing as scent camouflage.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 도축장 안, 양동이에 가득 담긴 검붉은 염소 피를 자신의 머리 위로 쏟아붓는 도중인 강민숙(홍은미)의 젖은 상체.\n\nLOCATION (lock): Inside the dark slaughterhouse beside buckets of collected goat blood. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook becomes completely coated in fresh goat blood, which must remain on her body and clothing as scent camouflage.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S77sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S77sh1.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "도축장이 너무 밝게 묘사되었고 주변에 피가 담긴 양동이들이 부족하며, 무거운 양동이를 측면만 손바닥으로 누르듯 쥐고 있는 자세가 물리적으로 어색합니다." }, { "label": "B", "score": 9, "verdict_ko": "어두운 밤의 도축장 조명과 주변에 놓인 피 양동이들을 충실히 구현했으며, 피를 쏟아붓는 행동과 물리적 묘사가 훨씬 사실적입니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "도축장이 너무 밝게 묘사되었고 주변에 피가 담긴 양동이들이 부족하며, 무거운 양동이를 측면만 손바닥으로 누르듯 쥐고 있는 자세가 물리적으로 어색합니다." }, { "label": "B", "score": 9, "verdict_ko": "어두운 밤의 도축장 조명과 주변에 놓인 피 양동이들을 충실히 구현했으며, 피를 쏟아붓는 행동과 물리적 묘사가 훨씬 사실적입니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "머리에 얹은 손과 양동이를 잡고 있는 두 손까지 총 세 개의 손이 나타나는 치명적인 신체 구조 오류가 있어 실격입니다." }, { "label": "B", "score": 8, "verdict_ko": "지정된 도축장 배경과 야간 시간대를 잘 반영했으며, 인물이 스스로 피를 쏟아붓는 상체 샷의 요구사항을 해부학적 오류 없이 정확하게 구현했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 2, "verdict_ko": "머리에 얹은 손과 양동이를 잡고 있는 두 손까지 총 세 개의 손이 나타나는 치명적인 신체 구조 오류가 있어 실격입니다." }, { "label": "A", "score": 8, "verdict_ko": "지정된 도축장 배경과 야간 시간대를 잘 반영했으며, 인물이 스스로 피를 쏟아붓는 상체 샷의 요구사항을 해부학적 오류 없이 정확하게 구현했습니다." } ] }, "combined": { "totals": { "A": 15, "B": 11 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 15, "B": 11 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "도축장이 너무 밝게 묘사되었고 주변에 피가 담긴 양동이들이 부족하며, 무거운 양동이를 측면만 손바닥으로 누르듯 쥐고 있는 자세가 물리적으로 어색합니다." }, { "label": "B", "score": 9, "verdict_ko": "어두운 밤의 도축장 조명과 주변에 놓인 피 양동이들을 충실히 구현했으며, 피를 쏟아붓는 행동과 물리적 묘사가 훨씬 사실적입니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S77sh1__bgfirst_bg.png", "bg_asset_id": "40b4d9b1-cd1c-4960-b963-c912d4fe5fde", "bg_record_key": "S77sh1::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S77sh10": { "input_fingerprint": "6ef4d59844bdbd8f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 온몸에 붉은 피를 뒤집어쓴 채 너무 놀라 목석처럼 빳빳하게 굳어버린 정인우의 얼어붙은 전신.\n\nLOCATION (lock): The slaughterhouse work floor at night, among blood containers and butchering tools. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark slaughterhouse, buckets of goat blood, hanging implements, and grimy surfaces. Exclude the mother's pouring action and show the man frozen after being drenched in blood.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook, Inwoo, and Jihu are all drenched in goat blood, and the coating must persist as their scent camouflage. Jihu remains without her necklace; Inwoo keeps his backpack, cast, and concealed handgun.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 6, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 전신 샷과 피를 뒤집어쓴 상태를 잘 구현했으며, 프롬프트에 명시된 배낭이 가장 명확하게 묘사되었습니다." }, { "label": "A", "score": 6, "verdict_ko": "전신 샷과 장소의 연속성은 좋으나, B에 비해 배낭의 형태가 덜 뚜렷하고 바닥에 임의로 양동이가 추가되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "인물과 배경의 기본 조건은 충족했으나, 필수 요소인 배낭이 완전히 누락되어 감점되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S77sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S77sh1" } }, "S78sh1": { "input_fingerprint": "112924d82958dc71", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 일렁이는 붉은 모닥불 불빛과 기괴한 핏빛 바위를 배경으로 꼿꼿하게 서 있는 강영란(흰 머리 여자)의 압도적인 전신.\n\nLOCATION (lock): Before the blood-marked rock altar in the firelit forest ritual clearing. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same giant blood-streaked rock, ritual fires, smoke, and surrounding black forest. Exclude the bell-dancing action and show the elder standing commandingly before the man.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran remains in the red shaman's robe before the blood-marked rock, with Jihu's bone necklace retained at the altar.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 4, "B": 3, "C": 7 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "샷 텍스트에 명시되지 않은 인물이 포함되어 등장인물 제한 규정을 위반했으며, 요구된 단독 전신 프레이밍 대신 와이드 투샷을 렌더링함." }, { "label": "B", "score": 3, "verdict_ko": "텍스트에 없는 인물이 바닥에 누운 상태로 임의 추가되어 심각한 규정 위반이며, 프레이밍과 샷 크기 요구사항을 전혀 충족하지 못함." }, { "label": "C", "score": 7, "verdict_ko": "샷 텍스트가 요구한 '압도적인 전신' 단독 프레이밍과 등장인물 제한을 정확히 준수하며 '흰 머리' 묘사를 훌륭하게 구현함." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S73sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S73sh4" }, "lane_policy": "ab_select_bypass:prev" }, "S78sh4::bgfirst_bg": { "input_fingerprint": "d8a5ba6bffcfe0e1", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 등을 돌린 고산들을 향해 단호하게 한 손을 뻗은 강영란(흰 머리 여자)의 팔.\n\nLOCATION (lock): The edge of the ritual clearing beside the burning fires and great rock.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside Youngran, the camera finishes its inward move about one meter from her extended arm and looks along it toward Sandeul's turned back. Her hand and forearm occupy the near middle-left without exceeding the human-focused frame, while Sandeul remains smaller in the right background, preserving the physical line of interruption.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 핏빛 바위 (Blood-red in appearance); used as Partial background anchor behind Youngran, retaining continuity with the confrontation ground; 모닥불 (Lit and flickering); used as Background illumination behind the interruption line.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is an open ritual clearing enclosed on every side by forest. A towering giant rock occupies the rear of the principal ritual forecourt and faces into the open ground. The central altar stands immediately in front of the rock on the clearing side, creating the main spatial axis: outer clearing, altar, then giant rock. Open ground extends laterally around this forecourt and outward to the surrounding forest edge, allowing movement around the rock and assembly across the wider clearing. A traversable forest-side approach enters the outer clearing through the enclosing woodland, although the source does not establish its compass direction. No permanent fire structures are specified; bonfires, smoke, blood traces, restraints, offerings, equipment, and participant formations remain moment-specific additions rather than site markers.\n- clearing: forest-enclosed ritual clearing\n- rock: towering giant ritual rock\n- altar: central altar before the giant rock\n- boundary: surrounding forest edge\n- approach: traversable forest-side approach\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 등을 돌린 고산들을 향해 단호하게 한 손을 뻗은 강영란(흰 머리 여자)의 팔.\n\nLOCATION (lock): The edge of the ritual clearing beside the burning fires and great rock.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside Youngran, the camera finishes its inward move about one meter from her extended arm and looks along it toward Sandeul's turned back. Her hand and forearm occupy the near middle-left without exceeding the human-focused frame, while Sandeul remains smaller in the right background, preserving the physical line of interruption.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 핏빛 바위 (Blood-red in appearance); used as Partial background anchor behind Youngran, retaining continuity with the confrontation ground; 모닥불 (Lit and flickering); used as Background illumination behind the interruption line.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is an open ritual clearing enclosed on every side by forest. A towering giant rock occupies the rear of the principal ritual forecourt and faces into the open ground. The central altar stands immediately in front of the rock on the clearing side, creating the main spatial axis: outer clearing, altar, then giant rock. Open ground extends laterally around this forecourt and outward to the surrounding forest edge, allowing movement around the rock and assembly across the wider clearing. A traversable forest-side approach enters the outer clearing through the enclosing woodland, although the source does not establish its compass direction. No permanent fire structures are specified; bonfires, smoke, blood traces, restraints, offerings, equipment, and participant formations remain moment-specific additions rather than site markers.\n- clearing: forest-enclosed ritual clearing\n- rock: towering giant ritual rock\n- altar: central altar before the giant rock\n- boundary: surrounding forest edge\n- approach: traversable forest-side approach\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S78sh4__bgfirst_bg.png", "asset_id": "c0f0cd22-d77e-4243-a6bb-4bea5d442042", "input_asset_ids": [ "9988bbc9-56bd-4932-b09d-e6774a9e39ca" ] }, "S78sh4": { "input_fingerprint": "06a1a02941c509a3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 등을 돌린 고산들을 향해 단호하게 한 손을 뻗은 강영란(흰 머리 여자)의 팔.\n\nLOCATION (lock): The edge of the ritual clearing beside the burning fires and great rock. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same firelit rock altar, smoke, ritual clearing, and red-black palette. Exclude the elder's frontal stance and show her reaching after the man as he turns away.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe, and Jihu's necklace remains at the ritual site under her control.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 등을 돌린 고산들을 향해 단호하게 한 손을 뻗은 강영란(흰 머리 여자)의 팔.\n\nLOCATION (lock): The edge of the ritual clearing beside the burning fires and great rock. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe, and Jihu's necklace remains at the ritual site under her control.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 등을 돌린 고산들을 향해 단호하게 한 손을 뻗은 강영란(흰 머리 여자)의 팔.\n\nLOCATION (lock): The edge of the ritual clearing beside the burning fires and great rock. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe, and Jihu's necklace remains at the ritual site under her control.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 등을 돌린 고산들을 향해 단호하게 한 손을 뻗은 강영란(흰 머리 여자)의 팔.\n\nLOCATION (lock): The edge of the ritual clearing beside the burning fires and great rock. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe, and Jihu's necklace remains at the ritual site under her control.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S78sh4__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S78sh4__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S78sh4__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "B": 8, "A": 6, "C": 4 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "무당복과 모자 등 레퍼런스의 복장을 가장 정확히 재현했으며, 등을 돌린 인물을 향해 팔을 뻗는 동작과 구도도 완벽하게 연출되었습니다." }, { "label": "A", "score": 6, "verdict_ko": "행동과 구도는 제시문 및 레퍼런스와 일치하나, 캐릭터 레퍼런스에 있는 모자가 누락되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "팔을 뻗는 동작은 맞으나, 무당복의 화려한 금박 장식과 방울 등이 완전히 생략된 밋밋한 붉은 옷과 다른 형태의 모자가 렌더링되어 복장 고증에서 크게 감점됩니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S78sh4__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S78sh4__bgfirst_bg.png", "bg_asset_id": "c0f0cd22-d77e-4243-a6bb-4bea5d442042", "bg_record_key": "S78sh4::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S78sh1" } }, "S79sh1": { "input_fingerprint": "c6faf89c88b8059c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 환한 조명등이 켜져 대낮처럼 밝은 선착장에 정박해 있는 낡은 낚시 어선의 외부 전경.\n\nLOCATION (lock): The brightly floodlit island pier around a moored fishing boat at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hwang's fishing boat remains secured at the brightly lit dock under guard so the fugitives cannot use it.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 6, "A": 3 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "참조 이미지의 공간적 요소(배경 여객선 등)를 가장 충실히 재현했으며 텍스트 금지 지시를 완벽히 준수했습니다." }, { "label": "B", "score": 6, "verdict_ko": "대낮처럼 밝은 조명 연출은 우수하나, 참조 이미지에 존재하는 배경 여객선이 누락되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "프롬프트에서 명시적으로 금지한 텍스트(출입금지 표지판)가 눈에 띄게 생성되어 큰 감점 요인이 되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L19B01.png" } ], "critique_skipped": true, "ref_mode": "lane(map_marker): 스케치", "share_plan": { "ref_plan": "background" } }, "S79sh5": { "input_fingerprint": "7993484edaeceaf3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 플래시 불빛을 어둠 속으로 뻗은 채 좁고 어두운 마을 골목길 한곳을 주시하고 있는 한국인 마을 사람들의 긴박한 뒷모습.\n\nLOCATION (lock): A narrow village alley branching away from the illuminated island pier. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same brightly lit dock, moored fishing boat, village entrance, and deep surrounding darkness. Exclude the static boat-centered composition and show search parties entering the narrow lanes with flashlights.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The villagers continue carrying flashlights and portable lights while searching; Hwang's boat remains guarded at the dock.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "어둠 속 골목길을 향해 플래시를 비추는 마을 사람들의 뒷모습을 프롬프트의 지시대로 가장 충실하게 연출했습니다." }, { "label": "B", "score": 3, "verdict_ko": "좁은 골목길 바로 옆에 바다와 배가 위치하는 등 참조 이미지의 지리적 구조를 왜곡한 심각한 공간 오류가 있습니다." }, { "label": "C", "score": 3, "verdict_ko": "어두운 골목길 안쪽으로 진입하는 대신 밝게 빛나는 선착장을 향해 걷고 있어 동선과 카메라 방향이 완전히 잘못되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S79sh1_sel.png" }, { "label": "PROP REFERENCE — 휴대용 조명 도구: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S79sh1" } }, "S80sh6::bgfirst_bg": { "input_fingerprint": "e7ebb7f6bf86f3ae", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 거대한 그물 무더기 사이로 얼굴을 절반쯤 내민 채 굳어 있는 피투성이 김지후(강수리영)의 조심스러운 표정.\n\nLOCATION (lock): A concealed nook behind stacked fishing nets inside the dark gear warehouse beside the pier.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level, the camera continues its low lateral track roughly one meter from the opening in the piled nets, tightening onto Sooryeong's half-concealed face. The net opening frames one side of her bloodied face while stacked fishing gear remains visible behind her as scale and concealment context.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 겹겹이 쌓인 어망 (Piled in layers) — The camera sees Sooryeong through a narrow opening between the piled nets; used as Foreground concealment and irregular frame around half of Sooryeong's face; 낚싯대와 통발 (Stacked inside the warehouse) — Their stacked lengths and openings angle away behind Sooryeong; used as Dense background clutter that limits movement and reinforces hiding; 큼직한 생선 바구니 (Stacked high) — Their sides face the concealed corner in repeated stacked rows; used as Background scale reference showing the warehouse is crowded with equipment; 창고 문 (Closed) — Its inner face is visible from the concealed corner; used as Sealed background boundary after the searcher leaves.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 거대한 그물 무더기 사이로 얼굴을 절반쯤 내민 채 굳어 있는 피투성이 김지후(강수리영)의 조심스러운 표정.\n\nLOCATION (lock): A concealed nook behind stacked fishing nets inside the dark gear warehouse beside the pier.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level, the camera continues its low lateral track roughly one meter from the opening in the piled nets, tightening onto Sooryeong's half-concealed face. The net opening frames one side of her bloodied face while stacked fishing gear remains visible behind her as scale and concealment context.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 김지후(강수리영) in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 겹겹이 쌓인 어망 (Piled in layers) — The camera sees Sooryeong through a narrow opening between the piled nets; used as Foreground concealment and irregular frame around half of Sooryeong's face; 낚싯대와 통발 (Stacked inside the warehouse) — Their stacked lengths and openings angle away behind Sooryeong; used as Dense background clutter that limits movement and reinforces hiding; 큼직한 생선 바구니 (Stacked high) — Their sides face the concealed corner in repeated stacked rows; used as Background scale reference showing the warehouse is crowded with equipment; 창고 문 (Closed) — Its inner face is visible from the concealed corner; used as Sealed background boundary after the searcher leaves.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh6__bgfirst_bg.png", "asset_id": "d5eae059-62b7-4e70-ad16-9b4dcc0f0644", "input_asset_ids": [ "fda1bbe0-2bef-4412-9884-b71bbb767961", "22b20647-c0e6-4261-a1ac-54af46888499" ] }, "S80sh6": { "input_fingerprint": "bbc98b65d52579b9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 거대한 그물 무더기 사이로 얼굴을 절반쯤 내민 채 굳어 있는 피투성이 김지후(강수리영)의 조심스러운 표정.\n\nLOCATION (lock): A concealed nook behind stacked fishing nets inside the dark gear warehouse beside the pier. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain coated in goat blood while hiding among the fishing gear. Jihu lacks her necklace, and Inwoo retains his backpack, cast, and concealed revolver.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 거대한 그물 무더기 사이로 얼굴을 절반쯤 내민 채 굳어 있는 피투성이 김지후(강수리영)의 조심스러운 표정.\n\nLOCATION (lock): A concealed nook behind stacked fishing nets inside the dark gear warehouse beside the pier. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain coated in goat blood while hiding among the fishing gear. Jihu lacks her necklace, and Inwoo retains his backpack, cast, and concealed revolver.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 거대한 그물 무더기 사이로 얼굴을 절반쯤 내민 채 굳어 있는 피투성이 김지후(강수리영)의 조심스러운 표정.\n\nLOCATION (lock): A concealed nook behind stacked fishing nets inside the dark gear warehouse beside the pier. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain coated in goat blood while hiding among the fishing gear. Jihu lacks her necklace, and Inwoo retains his backpack, cast, and concealed revolver.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh6__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S80sh6.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L49B05.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "레퍼런스 공간의 그물 무더기 사이로 조심스럽게 얼굴을 내민 피투성이 캐릭터의 모습을 사실적이고 자연스럽게 연출했습니다." }, { "label": "A", "score": 6, "verdict_ko": "얼굴을 반쯤 가린 지시문은 따랐으나, 카메라 바로 앞을 그물로 인위적으로 둘러싼 구도가 라이브 액션 샷으로서의 현실성을 떨어뜨립니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "레퍼런스 공간의 그물 무더기 사이로 조심스럽게 얼굴을 내민 피투성이 캐릭터의 모습을 사실적이고 자연스럽게 연출했습니다." }, { "label": "A", "score": 6, "verdict_ko": "얼굴을 반쯤 가린 지시문은 따랐으나, 카메라 바로 앞을 그물로 인위적으로 둘러싼 구도가 라이브 액션 샷으로서의 현실성을 떨어뜨립니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시문이 요구한 '그물 사이로 반쯤 내민 얼굴'과 '조심스러운 표정'을 적절한 구도와 크기로 훌륭하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "구도가 너무 넓고 피사체가 작아 지시문의 핵심인 '표정'을 명확히 전달하지 못합니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시문이 요구한 '그물 사이로 반쯤 내민 얼굴'과 '조심스러운 표정'을 적절한 구도와 크기로 훌륭하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "구도가 너무 넓고 피사체가 작아 지시문의 핵심인 '표정'을 명확히 전달하지 못합니다." } ] }, "combined": { "totals": { "A": 10, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 10, "B": 15 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "레퍼런스 공간의 그물 무더기 사이로 조심스럽게 얼굴을 내민 피투성이 캐릭터의 모습을 사실적이고 자연스럽게 연출했습니다." }, { "label": "A", "score": 6, "verdict_ko": "얼굴을 반쯤 가린 지시문은 따랐으나, 카메라 바로 앞을 그물로 인위적으로 둘러싼 구도가 라이브 액션 샷으로서의 현실성을 떨어뜨립니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L49B05.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh6__bgfirst_bg.png", "bg_asset_id": "d5eae059-62b7-4e70-ad16-9b4dcc0f0644", "bg_record_key": "S80sh6::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S80sh14::bgfirst_bg": { "input_fingerprint": "83b7452b2599b166", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 초가집 부엌 바닥에서 피범벅이 된 손으로 시체의 살점을 쥔 채 입가에 피를 묻히고 웅크린 경신 대기근의 아낙과 어린 한국인 아이의 모습.\n\nLOCATION (lock): Inside the bare kitchen of a famine-era thatched cottage, dimly lit in daytime.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Near floor level, the dolly reaches its closest point within one meter and settles on the woman and child crouched beside the corpse. Their bloodied hands and the held flesh occupy the lower center, their faces remain above in the same focal plane, and only a partial section of the corpse enters along the frame edge to preserve spatial context.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 초가집 부엌 바닥 (Interior floor; no additional surface state specified); used as Low compositional plane binding the living figures to the corpse and their actions; 초가집 부엌 (Dark interior during daytime) — The camera faces into the kitchen from near floor level; used as Limited environmental context around the crouched group.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 초가집 부엌 바닥에서 피범벅이 된 손으로 시체의 살점을 쥔 채 입가에 피를 묻히고 웅크린 경신 대기근의 아낙과 어린 한국인 아이의 모습.\n\nLOCATION (lock): Inside the bare kitchen of a famine-era thatched cottage, dimly lit in daytime.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Near floor level, the dolly reaches its closest point within one meter and settles on the woman and child crouched beside the corpse. Their bloodied hands and the held flesh occupy the lower center, their faces remain above in the same focal plane, and only a partial section of the corpse enters along the frame edge to preserve spatial context.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 초가집 부엌 바닥 (Interior floor; no additional surface state specified); used as Low compositional plane binding the living figures to the corpse and their actions; 초가집 부엌 (Dark interior during daytime) — The camera faces into the kitchen from near floor level; used as Limited environmental context around the crouched group.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh14__bgfirst_bg.png", "asset_id": "909053a1-54cf-4205-a67b-86a379b4971e", "input_asset_ids": [ "599acdcf-66f7-4247-be9f-5fbc5d6b3077", "de6d3d56-a042-43c6-99c6-d624c4abbbcc" ] }, "S80sh14": { "input_fingerprint": "e9a73502a14ef086", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 초가집 부엌 바닥에서 피범벅이 된 손으로 시체의 살점을 쥔 채 입가에 피를 묻히고 웅크린 경신 대기근의 아낙과 어린 한국인 아이의 모습.\n\nLOCATION (lock): Inside the bare kitchen of a famine-era thatched cottage, dimly lit in daytime. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The dead husband and father is sprawled on the dirt kitchen floor directly before his wife and child, his body motionless as they tear flesh from the corpse.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The starving woman and child remain smeared with blood as they crouch beside and eat from the dead husband and father, who is sprawled on the dirt kitchen floor directly before his wife and child, his body motionless.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 경신 대기근의 아낙 (조선인 성인 여성, 야윈 얼굴, 흐트러진 긴 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 초가집 부엌 바닥에서 피범벅이 된 손으로 시체의 살점을 쥔 채 입가에 피를 묻히고 웅크린 경신 대기근의 아낙과 어린 한국인 아이의 모습.\n\nLOCATION (lock): Inside the bare kitchen of a famine-era thatched cottage, dimly lit in daytime. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The dead husband and father is sprawled on the dirt kitchen floor directly before his wife and child, his body motionless as they tear flesh from the corpse.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The starving woman and child remain smeared with blood as they crouch beside and eat from the dead husband and father, who is sprawled on the dirt kitchen floor directly before his wife and child, his body motionless.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 경신 대기근의 아낙 (조선인 성인 여성, 야윈 얼굴, 흐트러진 긴 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 초가집 부엌 바닥에서 피범벅이 된 손으로 시체의 살점을 쥔 채 입가에 피를 묻히고 웅크린 경신 대기근의 아낙과 어린 한국인 아이의 모습.\n\nLOCATION (lock): Inside the bare kitchen of a famine-era thatched cottage, dimly lit in daytime. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The dead husband and father is sprawled on the dirt kitchen floor directly before his wife and child, his body motionless as they tear flesh from the corpse.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The starving woman and child remain smeared with blood as they crouch beside and eat from the dead husband and father, who is sprawled on the dirt kitchen floor directly before his wife and child, his body motionless.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 경신 대기근의 아낙 (조선인 성인 여성, 야윈 얼굴, 흐트러진 긴 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh14__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S80sh14.png" }, { "label": "CHARACTER REFERENCE — 경신 대기근의 아낙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L49B01.png" }, { "label": "CHARACTER REFERENCE — 경신 대기근의 아낙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 3, "verdict_ko": "지정된 야간 시간대는 준수했으나, 지시문의 핵심인 시체, 피 묻은 손과 입가 등의 잔혹한 묘사가 완전히 누락되었습니다." }, { "label": "A", "score": 2, "verdict_ko": "핵심인 시체와 피범벅이 된 모습이 전혀 없으며, 시간대 지침(밤)을 어기고 문밖을 낮으로 묘사했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 3, "verdict_ko": "지정된 야간 시간대는 준수했으나, 지시문의 핵심인 시체, 피 묻은 손과 입가 등의 잔혹한 묘사가 완전히 누락되었습니다." }, { "label": "A", "score": 2, "verdict_ko": "핵심인 시체와 피범벅이 된 모습이 전혀 없으며, 시간대 지침(밤)을 어기고 문밖을 낮으로 묘사했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "밤 시간대 지침은 충족했으나, 프롬프트의 핵심인 시체, 피범벅이 된 손, 살점 등의 묘사가 완전히 누락되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "지정된 밤 시간대를 위반하여 문밖이 낮으로 묘사되었으며, 시체와 혈흔 등 주요 상황 묘사도 전혀 나타나지 않습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "밤 시간대 지침은 충족했으나, 프롬프트의 핵심인 시체, 피범벅이 된 손, 살점 등의 묘사가 완전히 누락되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "지정된 밤 시간대를 위반하여 문밖이 낮으로 묘사되었으며, 시체와 혈흔 등 주요 상황 묘사도 전혀 나타나지 않습니다." } ] }, "combined": { "totals": { "A": 5, "B": 7 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 5, "B": 7 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 3, "verdict_ko": "지정된 야간 시간대는 준수했으나, 지시문의 핵심인 시체, 피 묻은 손과 입가 등의 잔혹한 묘사가 완전히 누락되었습니다." }, { "label": "A", "score": 2, "verdict_ko": "핵심인 시체와 피범벅이 된 모습이 전혀 없으며, 시간대 지침(밤)을 어기고 문밖을 낮으로 묘사했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L49B01.png" }, { "label": "CHARACTER REFERENCE — 경신 대기근의 아낙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh14__bgfirst_bg.png", "bg_asset_id": "909053a1-54cf-4205-a67b-86a379b4971e", "bg_record_key": "S80sh14::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S80sh25::bgfirst_bg": { "input_fingerprint": "e1791dd4423bc76e", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 배에서 한 발을 뻗어 내딛는 중간 자세인 낯선 조선인들을 향해 환한 미소로 양팔을 벌린 채 서 있는 수십 명의 한국인 섬 원주민들의 전면.\n\nLOCATION (lock): A primitive island landing where local residents welcome newcomers stepping off a boat in daylight.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height at the arrival point, the static camera faces the frontal line of smiling island residents past the strangers stepping from the boat. The residents spread across the middle and background as the primary plane, while the strangers remain smaller at the lower frame edge with one foot caught between boat and shore.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 배 (Arriving at 금월도) — The disembarkation side faces the gathered residents and camera axis; used as Secondary arrival plane holding the strangers at its edge; 금월도 바닷가 (Daytime arrival point); used as Shared meeting ground between the boat and the welcoming island residents.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 배에서 한 발을 뻗어 내딛는 중간 자세인 낯선 조선인들을 향해 환한 미소로 양팔을 벌린 채 서 있는 수십 명의 한국인 섬 원주민들의 전면.\n\nLOCATION (lock): A primitive island landing where local residents welcome newcomers stepping off a boat in daylight.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height at the arrival point, the static camera faces the frontal line of smiling island residents past the strangers stepping from the boat. The residents spread across the middle and background as the primary plane, while the strangers remain smaller at the lower frame edge with one foot caught between boat and shore.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 배 (Arriving at 금월도) — The disembarkation side faces the gathered residents and camera axis; used as Secondary arrival plane holding the strangers at its edge; 금월도 바닷가 (Daytime arrival point); used as Shared meeting ground between the boat and the welcoming island residents.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh25__bgfirst_bg.png", "asset_id": "1775d41d-e05a-4d72-b0a4-5e3a57c36ca4", "input_asset_ids": [ "7d8e0879-b45e-4108-9f73-a0347a8fd8c3", "50b3c7b7-9895-48d5-8bbb-b60e84c7d62a" ] }, "S80sh25": { "input_fingerprint": "574ae9a1b3af98b3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 배에서 한 발을 뻗어 내딛는 중간 자세인 낯선 조선인들을 향해 환한 미소로 양팔을 벌린 채 서 있는 수십 명의 한국인 섬 원주민들의 전면.\n\nLOCATION (lock): A primitive island landing where local residents welcome newcomers stepping off a boat in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The arriving flesh-eaters retain their carried belongings as they step off the boat; the welcoming islanders are still alive at this point in the historical montage.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 배에서 한 발을 뻗어 내딛는 중간 자세인 낯선 조선인들을 향해 환한 미소로 양팔을 벌린 채 서 있는 수십 명의 한국인 섬 원주민들의 전면.\n\nLOCATION (lock): A primitive island landing where local residents welcome newcomers stepping off a boat in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The arriving flesh-eaters retain their carried belongings as they step off the boat; the welcoming islanders are still alive at this point in the historical montage.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 배에서 한 발을 뻗어 내딛는 중간 자세인 낯선 조선인들을 향해 환한 미소로 양팔을 벌린 채 서 있는 수십 명의 한국인 섬 원주민들의 전면.\n\nLOCATION (lock): A primitive island landing where local residents welcome newcomers stepping off a boat in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The arriving flesh-eaters retain their carried belongings as they step off the boat; the welcoming islanders are still alive at this point in the historical montage.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh25__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S80sh25.png" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L49B02.png" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "섬 원주민들의 환한 미소와 앞모습(전면)을 잘 포착했으며, 배에서 한 발을 내딛는 동작과 소지품 유지 조건도 훌륭하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "원주민들의 앞모습(전면)보다는 측면/뒷모습이 많이 보이며, 미소보다는 부자연스러운 자세가 두드러져 지시문 구현도가 떨어집니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "섬 원주민들의 환한 미소와 앞모습(전면)을 잘 포착했으며, 배에서 한 발을 내딛는 동작과 소지품 유지 조건도 훌륭하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "원주민들의 앞모습(전면)보다는 측면/뒷모습이 많이 보이며, 미소보다는 부자연스러운 자세가 두드러져 지시문 구현도가 떨어집니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 행동(배에서 발을 내딛는 동작, 미소 지으며 팔을 벌린 원주민들)과 카메라 앵글(원주민들의 전면)을 잘 구현했으며 레퍼런스 구도 복사 금지 조건도 준수했습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스 이미지의 카메라 구도를 그대로 복사하지 말라는 지시를 위반했으며, 원주민들의 표정과 자세가 부자연스럽습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 행동(배에서 발을 내딛는 동작, 미소 지으며 팔을 벌린 원주민들)과 카메라 앵글(원주민들의 전면)을 잘 구현했으며 레퍼런스 구도 복사 금지 조건도 준수했습니다." }, { "label": "A", "score": 3, "verdict_ko": "레퍼런스 이미지의 카메라 구도를 그대로 복사하지 말라는 지시를 위반했으며, 원주민들의 표정과 자세가 부자연스럽습니다." } ] }, "combined": { "totals": { "A": 7, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 7, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "섬 원주민들의 환한 미소와 앞모습(전면)을 잘 포착했으며, 배에서 한 발을 내딛는 동작과 소지품 유지 조건도 훌륭하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "원주민들의 앞모습(전면)보다는 측면/뒷모습이 많이 보이며, 미소보다는 부자연스러운 자세가 두드러져 지시문 구현도가 떨어집니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L49B02.png" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh25__bgfirst_bg.png", "bg_asset_id": "1775d41d-e05a-4d72-b0a4-5e3a57c36ca4", "bg_record_key": "S80sh25::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S81sh9": { "input_fingerprint": "f9311edf6abc5796", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 한쪽 소매를 눈가에 가져다 댄 채 원망 섞인 표정으로 입을 열고 있는 김지후(강수리영)의 상체.\n\nLOCATION (lock): Among stacked nets and fishing gear inside the dim pier-side warehouse. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same piled nets, fishing rods, traps, baskets, and dim warehouse light. Exclude the immediate hiding-search action and show the daughter confronting her mother among the gear.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains coated in dried goat blood and without her necklace; Inwoo and Minsook retain the same blood camouflage, while Inwoo keeps his backpack, cast, and concealed revolver.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 6, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "소매를 눈가에 가져다 대고 원망 섞인 표정으로 입을 여는 동작을 자연스럽게 연출했으며, 지정된 단일 인물 조건과 배경을 정확히 준수했습니다." }, { "label": "A", "score": 6, "verdict_ko": "샷 텍스트의 요구 사항(상체, 표정, 입 열림)을 잘 구현했으나, 소매가 눈가에 닿는 디테일이 약간 어색합니다." }, { "label": "C", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 다른 인물들을 배경에 임의로 추가하여 심각한 지침 위반(Hard Violation)에 해당합니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S80sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S80sh6" } }, "S81sh12": { "input_fingerprint": "68aa9d54eede31ba", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 참지 못하고 오열하며 두 손으로 얼굴을 감싼 강민숙(홍은미)의 무너진 상체.\n\nLOCATION (lock): The concealed rear corner of the pier-side gear warehouse. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cramped fishing-gear stacks, dusty darkness, and concealed corner. Exclude the searching youth and show the mother collapsing into grief.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains coated in goat blood while breaking down in tears; Jihu and Inwoo retain the same camouflage nearby.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 4, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "두 손으로 얼굴을 감싸며 무너진 상체를 원본 배경과 일관되게, 해부학적 오류 없이 가장 정확하게 묘사했습니다." }, { "label": "C", "score": 4, "verdict_ko": "이전 샷에 없던 어구 상자가 피사체 바로 앞에 무단으로 추가되어 공간의 연속성을 훼손했습니다." }, { "label": "A", "score": 3, "verdict_ko": "손가락 개수가 비정상적으로 묘사되는 심각한 해부학적 오류(Hard Violation)가 발생했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S81sh9_sel.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S81sh9" } }, "S81sh13": { "input_fingerprint": "a21ce9c462f0ebcc", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 눈물을 흘리면서도 떨리는 눈빛으로 오열하는 엄마를 묵묵히 응시하는 김지후(강수리영)의 굳은 옆모습.\n\nLOCATION (lock): Inside the cramped hiding place behind fishing equipment in the gear warehouse. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same nets, stacked equipment, narrow floor space, and dim night light. Exclude the mother's face-covering pose from the foreground and frame the daughter watching her in tears.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains goat-blood-covered and necklace-less as she watches her mother; Inwoo and Minsook remain similarly coated.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 9, "C": 7, "A": 5 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "눈물을 흘리는 지후의 굳은 옆모습과 배경에서 오열하는 엄마의 구도가 완벽하며, 목걸이가 없는 상태도 정확히 반영되었습니다." }, { "label": "C", "score": 7, "verdict_ko": "지후의 굳은 옆모습과 배경의 엄마는 잘 연출되었으나, 지후가 눈물을 흘리는 디테일이 생략되었습니다." }, { "label": "A", "score": 5, "verdict_ko": "지후가 눈물을 흘리지 않으며, 지시문에서 제외를 요구한 목걸이 줄이 남아 있어 감점되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S81sh12_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S81sh12" } }, "S82sh2": { "input_fingerprint": "46547a2c9a537b7b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): dawn transitioning to morning.\n\nSHOT TEXT (authoritative, Korean): 수평선 너머로 눈부신 아침 해가 반쯤 모습을 드러낸 장엄한 일출 전경.\n\nLOCATION (lock): The open sea and horizon directly off the island’s coastal village at sunrise. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 8, "B": 6, "C": 5 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "수평선 너머로 태양이 반쯤 드러난 모습을 프롬프트 지시대로 가장 정확하게 연출함." }, { "label": "B", "score": 6, "verdict_ko": "아름다운 일출 풍경이나, 태양이 수평선 위로 완전히 떠올라 '반쯤 모습을 드러낸'이라는 묘사와 어긋남." }, { "label": "C", "score": 5, "verdict_ko": "태양이 수평선에서 완전히 떨어져 떠오른 상태로, 프롬프트의 구체적인 시간대 묘사를 놓침." } ], "refs": [], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S83sh1::bgfirst_bg": { "input_fingerprint": "b255f8603573e1fe", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 식당 테이블에 앉아 열린 문 너머 바깥을 뚫어지게 응시하는 강영란(흰 머리 여자)의 굳은 상체.\n\nLOCATION (lock): At a table inside the restaurant hall, facing the open doorway and daylight village beyond.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From seated eye height inside the dining room, frame 강영란(흰 머리 여자) in a tight medium three-quarter view at the table, with the open doorway retained near the opposite edge along her gaze axis. Her rigid torso occupies the left-middle of frame while the doorway provides a clear visual route toward the activity outside without weakening her fixed concentration.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-left of the frame, midground, looks toward open doorway and search outside; open doorway in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 열린 식당 문 (Open) — The opening faces the camera obliquely, revealing the exterior search area beyond; used as Maintained as the directional endpoint of 강영란(흰 머리 여자)'s stare and the next pan axis; 식당 테이블 (In use as seating for 강영란(흰 머리 여자)) — Its near corner enters the lower frame, with 강영란(흰 머리 여자) seated along its far side; used as Grounds her seated position and marks the place usually occupied by 수리영; 수색 중인 마을 주민들 (Moving back and forth while searching); used as Visible through the doorway as small moving figures that contextualize her vigilance.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 식당 테이블에 앉아 열린 문 너머 바깥을 뚫어지게 응시하는 강영란(흰 머리 여자)의 굳은 상체.\n\nLOCATION (lock): At a table inside the restaurant hall, facing the open doorway and daylight village beyond.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From seated eye height inside the dining room, frame 강영란(흰 머리 여자) in a tight medium three-quarter view at the table, with the open doorway retained near the opposite edge along her gaze axis. Her rigid torso occupies the left-middle of frame while the doorway provides a clear visual route toward the activity outside without weakening her fixed concentration.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-left of the frame, midground, looks toward open doorway and search outside; open doorway in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 열린 식당 문 (Open) — The opening faces the camera obliquely, revealing the exterior search area beyond; used as Maintained as the directional endpoint of 강영란(흰 머리 여자)'s stare and the next pan axis; 식당 테이블 (In use as seating for 강영란(흰 머리 여자)) — Its near corner enters the lower frame, with 강영란(흰 머리 여자) seated along its far side; used as Grounds her seated position and marks the place usually occupied by 수리영; 수색 중인 마을 주민들 (Moving back and forth while searching); used as Visible through the doorway as small moving figures that contextualize her vigilance.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S83sh1__bgfirst_bg.png", "asset_id": "4ca48fdc-7216-49ff-9007-b92aee39cc04", "input_asset_ids": [ "44860d32-ce3a-40f6-bfd4-fa25aa1a4d3e", "15c006bc-c7ae-4006-8e15-ae5074040f95" ] }, "S83sh1": { "input_fingerprint": "7f7cb4775ad35a7b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 식당 테이블에 앉아 열린 문 너머 바깥을 뚫어지게 응시하는 강영란(흰 머리 여자)의 굳은 상체.\n\nLOCATION (lock): At a table inside the restaurant hall, facing the open doorway and daylight village beyond. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rear shed remains a burned ruin outside, and Jihu's bone necklace remains in Youngran's possession at the ritual site.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 식당 테이블에 앉아 열린 문 너머 바깥을 뚫어지게 응시하는 강영란(흰 머리 여자)의 굳은 상체.\n\nLOCATION (lock): At a table inside the restaurant hall, facing the open doorway and daylight village beyond. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rear shed remains a burned ruin outside, and Jihu's bone necklace remains in Youngran's possession at the ritual site.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 식당 테이블에 앉아 열린 문 너머 바깥을 뚫어지게 응시하는 강영란(흰 머리 여자)의 굳은 상체.\n\nLOCATION (lock): At a table inside the restaurant hall, facing the open doorway and daylight village beyond. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rear shed remains a burned ruin outside, and Jihu's bone necklace remains in Youngran's possession at the ritual site.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S83sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S83sh1.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B06.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "지문에 없는 다수의 인물들이 문 밖에 임의로 추가되어 심각한 규칙 위반(Hard Violation)이며, 불탄 헛간의 연속성 설정도 누락되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "테이블에 앉아 문 밖을 응시하는 인물의 상체를 정확한 구도로 연출했으며, 문 밖의 불탄 헛간 연속성 설정도 충실히 반영했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "지문에 없는 다수의 인물들이 문 밖에 임의로 추가되어 심각한 규칙 위반(Hard Violation)이며, 불탄 헛간의 연속성 설정도 누락되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "테이블에 앉아 문 밖을 응시하는 인물의 상체를 정확한 구도로 연출했으며, 문 밖의 불탄 헛간 연속성 설정도 충실히 반영했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 위치와 인물 외형을 잘 반영했으며, 문 밖의 불탄 헛간 구조물을 정확히 구현하고 불필요한 인물을 추가하지 않아 규정을 준수했습니다." }, { "label": "B", "score": 3, "verdict_ko": "문 밖 풍경에 프롬프트에 명시되지 않은 여러 명의 사람들을 임의로 추가하여 심각한 규정 위반(Hard Violation)이 발생했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 위치와 인물 외형을 잘 반영했으며, 문 밖의 불탄 헛간 구조물을 정확히 구현하고 불필요한 인물을 추가하지 않아 규정을 준수했습니다." }, { "label": "A", "score": 3, "verdict_ko": "문 밖 풍경에 프롬프트에 명시되지 않은 여러 명의 사람들을 임의로 추가하여 심각한 규정 위반(Hard Violation)이 발생했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "지문에 없는 다수의 인물들이 문 밖에 임의로 추가되어 심각한 규칙 위반(Hard Violation)이며, 불탄 헛간의 연속성 설정도 누락되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "테이블에 앉아 문 밖을 응시하는 인물의 상체를 정확한 구도로 연출했으며, 문 밖의 불탄 헛간 연속성 설정도 충실히 반영했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B06.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S83sh1__bgfirst_bg.png", "bg_asset_id": "4ca48fdc-7216-49ff-9007-b92aee39cc04", "bg_record_key": "S83sh1::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S83sh3": { "input_fingerprint": "992ebc1f2a9f6a9f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 문가에 우두커니 선 채 강영란(흰 머리 여자)을 쳐다보는 고산들의 굳건한 전면.\n\nLOCATION (lock): The interior side of the restaurant entrance, overlooking the village and sea. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same restaurant table, open doorway, daylight, and view toward the harbor. Exclude the elder's solitary seated emphasis and add the man standing at the door watching her.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu's bone necklace remains under Youngran's control, while the burned shed remains destroyed behind the restaurant.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 4, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "샷 텍스트에 명시된 고산들의 정면과 시선 교환을 완벽히 연출했으며, 이전 컷의 공간 구조와 인물의 위치를 충실히 유지했습니다." }, { "label": "A", "score": 4, "verdict_ko": "주요 피사체인 흰 머리 여자가 프레임에서 누락되었으며, 인물 간의 시선과 상황이 제시된 샷 텍스트와 일치하지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "공간의 문과 창문 구조가 레퍼런스와 다르게 변형되었고, 남성이 모자를 쓰고 있어 '짧게 깎은 머리'라는 외형 지시를 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S83sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S83sh1" } }, "S84sh3::bgfirst_bg": { "input_fingerprint": "569a60d37d6b45f1", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 백팩 안에서 38구경 권총을 반쯤 꺼낸 채 멈춘 정인우의 긴장된 손.\n\nLOCATION (lock): The hidden corner of the pier-side gear warehouse among stacked nets and traps.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the downward tilt at hand height beside 정인우, very close to the backpack opening, isolating his tense hand and the half-drawn .38 caliber handgun while retaining a narrow strip of bag and clothing for scale. The grip and hand occupy the central lower area without exceeding a natural proportion, and the halted extraction reads as the instant before a dangerous decision.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: .38 caliber handgun in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 백팩 입구 (Open while the handgun is being drawn) — The opening angles toward the camera beside 정인우's hand; used as Immediate frame around the hand and weapon, providing scale and concealment context; 38구경 권총 (Half drawn and held still) — Its grip and part of the weapon face the camera while the remainder stays inside the backpack; used as Primary detail held between concealment and full exposure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 백팩 안에서 38구경 권총을 반쯤 꺼낸 채 멈춘 정인우의 긴장된 손.\n\nLOCATION (lock): The hidden corner of the pier-side gear warehouse among stacked nets and traps.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the downward tilt at hand height beside 정인우, very close to the backpack opening, isolating his tense hand and the half-drawn .38 caliber handgun while retaining a narrow strip of bag and clothing for scale. The grip and hand occupy the central lower area without exceeding a natural proportion, and the halted extraction reads as the instant before a dangerous decision.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: .38 caliber handgun in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 백팩 입구 (Open while the handgun is being drawn) — The opening angles toward the camera beside 정인우's hand; used as Immediate frame around the hand and weapon, providing scale and concealment context; 38구경 권총 (Half drawn and held still) — Its grip and part of the weapon face the camera while the remainder stays inside the backpack; used as Primary detail held between concealment and full exposure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S84sh3__bgfirst_bg.png", "asset_id": "418a0219-66e9-4575-adc2-690ab48d5001", "input_asset_ids": [ "fb7ea541-5a71-44f0-b51a-c18f3729ca10", "e8c14b89-925f-42b4-8d4e-cc3a3a86f891" ] }, "S84sh3": { "input_fingerprint": "38e6d6efc487a561", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 백팩 안에서 38구경 권총을 반쯤 꺼낸 채 멈춘 정인우의 긴장된 손.\n\nLOCATION (lock): The hidden corner of the pier-side gear warehouse among stacked nets and traps. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo possesses the .38-caliber revolver taken from his backpack and keeps the backpack with him; his injured arm remains cast. All three fugitives remain goat-blood-covered, and Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 백팩 안에서 38구경 권총을 반쯤 꺼낸 채 멈춘 정인우의 긴장된 손.\n\nLOCATION (lock): The hidden corner of the pier-side gear warehouse among stacked nets and traps. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo possesses the .38-caliber revolver taken from his backpack and keeps the backpack with him; his injured arm remains cast. All three fugitives remain goat-blood-covered, and Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 백팩 안에서 38구경 권총을 반쯤 꺼낸 채 멈춘 정인우의 긴장된 손.\n\nLOCATION (lock): The hidden corner of the pier-side gear warehouse among stacked nets and traps. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo possesses the .38-caliber revolver taken from his backpack and keeps the backpack with him; his injured arm remains cast. All three fugitives remain goat-blood-covered, and Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S84sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S84sh3.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 38구경 권총: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L49B03.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 38구경 권총: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "손의 해부학적 구조가 매우 부자연스러우며, 권총의 총열 길이가 레퍼런스와 다르게 왜곡된 하드 위반입니다." }, { "label": "B", "score": 7, "verdict_ko": "백팩에서 권총을 반쯤 꺼낸 긴장된 손의 묘사가 지문과 완벽히 일치하며, 깁스한 팔과 피 묻은 디테일, 소품 고증까지 충실히 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "손의 해부학적 구조가 매우 부자연스러우며, 권총의 총열 길이가 레퍼런스와 다르게 왜곡된 하드 위반입니다." }, { "label": "B", "score": 7, "verdict_ko": "백팩에서 권총을 반쯤 꺼낸 긴장된 손의 묘사가 지문과 완벽히 일치하며, 깁스한 팔과 피 묻은 디테일, 소품 고증까지 충실히 구현했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 '긴장된 손'과 '반쯤 꺼낸 총'에 온전히 집중하는 완벽한 클로즈업 프레이밍을 보여주며, 피 묻은 깁스와 배경 디테일까지 정확합니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 샷 텍스트의 프레이밍보다 불필요하게 넓은 앵글을 잡아 감점 요소가 되며, 총을 꺼내는 동작과 손의 형태가 다소 어색합니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 '긴장된 손'과 '반쯤 꺼낸 총'에 온전히 집중하는 완벽한 클로즈업 프레이밍을 보여주며, 피 묻은 깁스와 배경 디테일까지 정확합니다." }, { "label": "B", "score": 4, "verdict_ko": "지정된 샷 텍스트의 프레이밍보다 불필요하게 넓은 앵글을 잡아 감점 요소가 되며, 총을 꺼내는 동작과 손의 형태가 다소 어색합니다." } ] }, "combined": { "totals": { "A": 10, "B": 11 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 } }, "totals": { "A": 10, "B": 11 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "손의 해부학적 구조가 매우 부자연스러우며, 권총의 총열 길이가 레퍼런스와 다르게 왜곡된 하드 위반입니다." }, { "label": "B", "score": 7, "verdict_ko": "백팩에서 권총을 반쯤 꺼낸 긴장된 손의 묘사가 지문과 완벽히 일치하며, 깁스한 팔과 피 묻은 디테일, 소품 고증까지 충실히 구현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L49B03.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 38구경 권총: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S84sh3__bgfirst_bg.png", "bg_asset_id": "418a0219-66e9-4575-adc2-690ab48d5001", "bg_record_key": "S84sh3::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S84sh6": { "input_fingerprint": "fd43d1b04efe0b4f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 입을 굳게 다문 채 참담한 표정을 짓고 있는 강민숙(홍은미)의 굳은 얼굴.\n\nLOCATION (lock): Inside the dim fishing-gear warehouse near its dusty window overlooking the pier. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook, Jihu, and Inwoo remain coated in goat blood inside the fishing-gear shed. Inwoo retains the revolver and backpack, and Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 8, "A": 7, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "지문이 요구한 참담한 표정의 단독 클로즈업을 잘 구현했으며, 레퍼런스의 배경 구조(우측 창문, 좌측 그물)를 가장 정확하게 유지했습니다." }, { "label": "A", "score": 7, "verdict_ko": "인물의 표정과 프레이밍은 지문에 부합하나, 창문의 위치와 초록색 상자 등 배경 요소가 레퍼런스와 일치하지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 인물들(배경의 남성들)을 프레임 안에 임의로 추가하여 엄격한 인물 제한 지침을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S84sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S84sh3" } }, "S85sh2::bgfirst_bg": { "input_fingerprint": "288cad7f9823c531", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 몸의 방향은 숲을 향한 채 고개만 선착장 옆 낡은 어구 창고로 돌려 시선을 고정한 고산들의 상체.\n\nLOCATION (lock): A village lane near the public-address speaker tower, with a clear view toward the pier-side warehouse.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track at chest height beside and slightly ahead of 고산들, holding a medium upper-body composition that preserves the contradiction between his forest-facing torso and his head turned toward the old gear warehouse. Place him on the middle-left, leaving open space on the right along his fixed warehouse-directed gaze so the suspicion reads geometrically before the camera moves closer.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 고산들 in the middle-left of the frame, midground, looks toward old gear warehouse beside the dock; old gear warehouse beside the dock in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 선착장 옆 낡은 어구 창고 (Described as old; other conditions are not specified) — The warehouse is presented from the side facing 고산들 and the camera's gaze corridor; used as Required endpoint of 고산들's redirected attention, held across the open side of the frame; 숲 방향 (Search route toward which he had been moving); used as The direction his torso still favors, clarifying that only his attention has shifted.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nArrange the site as a connected inland-to-waterfront village edge. The village public-address loudspeaker installation occupies the village corner. The forest edge borders the uphill side of this sector, while the dockside waterfront lies in the opposite direction and remains visible when turning toward the forest. At the waterfront, the fishing-gear warehouse stands immediately beside the village dock, with its required entrance facing the accessible dockside area. The dock directly meets the sea, and the open sea approach aligns toward the dock. This creates a clear spatial sequence from sea to dock and warehouse, through the village loudspeaker area, and uphill to the forest edge.\n- loudspeaker: village public-address loudspeaker installation\n- forest: uphill forest edge\n- warehouse: dockside fishing-gear warehouse\n- entrance: dockside fishing-gear warehouse entrance\n- dock: village waterfront dock\n- sea: sea approach off the village dock\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 몸의 방향은 숲을 향한 채 고개만 선착장 옆 낡은 어구 창고로 돌려 시선을 고정한 고산들의 상체.\n\nLOCATION (lock): A village lane near the public-address speaker tower, with a clear view toward the pier-side warehouse.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track at chest height beside and slightly ahead of 고산들, holding a medium upper-body composition that preserves the contradiction between his forest-facing torso and his head turned toward the old gear warehouse. Place him on the middle-left, leaving open space on the right along his fixed warehouse-directed gaze so the suspicion reads geometrically before the camera moves closer.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 고산들 in the middle-left of the frame, midground, looks toward old gear warehouse beside the dock; old gear warehouse beside the dock in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 선착장 옆 낡은 어구 창고 (Described as old; other conditions are not specified) — The warehouse is presented from the side facing 고산들 and the camera's gaze corridor; used as Required endpoint of 고산들's redirected attention, held across the open side of the frame; 숲 방향 (Search route toward which he had been moving); used as The direction his torso still favors, clarifying that only his attention has shifted.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nArrange the site as a connected inland-to-waterfront village edge. The village public-address loudspeaker installation occupies the village corner. The forest edge borders the uphill side of this sector, while the dockside waterfront lies in the opposite direction and remains visible when turning toward the forest. At the waterfront, the fishing-gear warehouse stands immediately beside the village dock, with its required entrance facing the accessible dockside area. The dock directly meets the sea, and the open sea approach aligns toward the dock. This creates a clear spatial sequence from sea to dock and warehouse, through the village loudspeaker area, and uphill to the forest edge.\n- loudspeaker: village public-address loudspeaker installation\n- forest: uphill forest edge\n- warehouse: dockside fishing-gear warehouse\n- entrance: dockside fishing-gear warehouse entrance\n- dock: village waterfront dock\n- sea: sea approach off the village dock\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S85sh2__bgfirst_bg.png", "asset_id": "3c5cecb8-fd85-4557-906a-26c5029e18cd", "input_asset_ids": [ "67120033-305c-46d2-b2d9-0d446b13195a" ] }, "S85sh2": { "input_fingerprint": "f4e0377ba2f38a2d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 몸의 방향은 숲을 향한 채 고개만 선착장 옆 낡은 어구 창고로 돌려 시선을 고정한 고산들의 상체.\n\nLOCATION (lock): A village lane near the public-address speaker tower, with a clear view toward the pier-side warehouse. The shot takes place here — the attached STORYBOARD SKETCH fixes the staging, camera and figure placement of this exact place. No location photograph is attached — build the location itself strictly from the location text above and the shot text, inventing nothing beyond them.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain hidden inside the fishing-gear shed under their goat-blood camouflage; Inwoo retains the revolver and backpack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 몸의 방향은 숲을 향한 채 고개만 선착장 옆 낡은 어구 창고로 돌려 시선을 고정한 고산들의 상체.\n\nLOCATION (lock): A village lane near the public-address speaker tower, with a clear view toward the pier-side warehouse. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain hidden inside the fishing-gear shed under their goat-blood camouflage; Inwoo retains the revolver and backpack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 몸의 방향은 숲을 향한 채 고개만 선착장 옆 낡은 어구 창고로 돌려 시선을 고정한 고산들의 상체.\n\nLOCATION (lock): A village lane near the public-address speaker tower, with a clear view toward the pier-side warehouse. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain hidden inside the fishing-gear shed under their goat-blood camouflage; Inwoo retains the revolver and backpack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 몸의 방향은 숲을 향한 채 고개만 선착장 옆 낡은 어구 창고로 돌려 시선을 고정한 고산들의 상체.\n\nLOCATION (lock): A village lane near the public-address speaker tower, with a clear view toward the pier-side warehouse. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain hidden inside the fishing-gear shed under their goat-blood camouflage; Inwoo retains the revolver and backpack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S85sh2__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S85sh2__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S85sh2__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "C": 7, "A": 4, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "몸을 숲으로 향한 채 고개만 창고 쪽으로 돌린 지시문의 복잡한 자세를 완벽하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "몸의 방향이 숲을 향하지 않고 정면을 향하고 있어 샷의 핵심 자세 지시를 놓쳤습니다." }, { "label": "B", "score": 3, "verdict_ko": "자세가 지시문과 불일치하며, 지정된 짧은 머리 스타일(짧게 깎은 머리)도 반영되지 않았습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S85sh2__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S85sh2__bgfirst_bg.png", "bg_asset_id": "3c5cecb8-fd85-4557-906a-26c5029e18cd", "bg_record_key": "S85sh2::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+엔티티", "share_plan": { "ref_plan": "background" } }, "S85sh5": { "input_fingerprint": "a9990bb5849ff6c7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 뱃머리에 거친 물보라를 일으키며 금월도 선착장을 향해 돌진하는 순간에 포착된 경찰 순찰정의 전경.\n\nLOCATION (lock): The water approach immediately outside the island pier in daylight. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The approaching patrol boat remains headed directly toward the Geumwoldo dock, offering the hidden fugitives a possible escape.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 4, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "선착장으로 향하는 순찰정의 방향과 물보라를 정확히 묘사했으며, 레퍼런스의 배경 구조물(흰 건물, 스피커 폴대)을 가장 충실하게 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "배의 진행 방향은 맞으나, 레퍼런스에 전혀 없는 마을과 산 지형이 배경에 임의로 추가되어 장소 일관성(Location Lock)을 크게 훼손했습니다." }, { "label": "C", "score": 3, "verdict_ko": "배경 장소는 완벽하게 구현했으나, 선착장으로 향해야 할 배가 정반대로 선착장을 등지고 멀어지는 방향으로 이동하고 있어 치명적인 방향 오류(Hard Violation)입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S85sh2_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S85sh2" }, "lane_policy": "share_plan_prev_bgonly" }, "S86sh5::bgfirst_bg": { "input_fingerprint": "fce65d0022206b84", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 선착장 위쪽에서 수리영 일행을 향해 몸을 굽히고 한 발을 뻗은 채 맹렬하게 뛰어내려오는 수많은 한국인 마을 사람들의 mid-action 전경.\n\nLOCATION (lock): The sloping village approach descending onto the island pier.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From slightly above head height near the lower dock, crane over a wide view looking back toward the upper side as the mass of Korean villagers pours downward. Arrange the leading bodies across the middle band with bent torsos and extended strides, while successive ranks climb into the upper background so the charge reads as a single predatory flow toward the unseen patrol boat below frame.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: charging crowd of Korean villagers in the upper-center of the frame, midground, moves toward lower dock and patrol boat; route toward the lower dock in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 선착장 위쪽에서 내려오는 길 (Occupied by villagers running toward the patrol boat) — Its downward axis runs from the upper background toward the lower center of frame; used as Descending route that organizes the crowd into depth and directs the charge toward camera; 선착장 하단 (The villagers are rapidly closing on it) — The approach side lies below the charging crowd and toward the camera; used as Lower spatial destination positioned close to the camera side of the action.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe main pier extends seaward from the village-side approach and terminates at an outer pier end surrounded by the adjacent sea. Its main deck provides a continuous route between the village and the outer end, while a vessel berth edge directly contacts the water along the boat-access side. At the landward end, the approach continues into the small island village. The fishing-gear warehouse and Banwol Restaurant face this pier-to-village route, with the warehouse doorway opening onto the landing-side escape path and the restaurant entrance addressing the same general frontage. Small village houses occupy the inland area around the approach, with woodland interwoven among and behind them. This arrangement allows views inward from the outer pier toward the village, outward from the village frontage toward the berth and sea, and lateral coverage between the warehouse, restaurant frontage, and main pier deck.\n- pier end: seaward terminal end of the island pier\n- pier: main island pier deck\n- berth: water-contacting vessel berth edge\n- water: sea adjoining the island pier\n- approach: village-to-pier approach\n- houses: small island village houses around the pier approach\n- woodland: woodland surrounding the small island village\n- warehouse: fishing-gear warehouse beside the pier landing\n- door: pier-facing fishing-gear warehouse doorway\n- restaurant: Banwol Restaurant at the village-side pier frontage\n- entrance: pier-visible entrance of Banwol Restaurant\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 선착장 위쪽에서 수리영 일행을 향해 몸을 굽히고 한 발을 뻗은 채 맹렬하게 뛰어내려오는 수많은 한국인 마을 사람들의 mid-action 전경.\n\nLOCATION (lock): The sloping village approach descending onto the island pier.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From slightly above head height near the lower dock, crane over a wide view looking back toward the upper side as the mass of Korean villagers pours downward. Arrange the leading bodies across the middle band with bent torsos and extended strides, while successive ranks climb into the upper background so the charge reads as a single predatory flow toward the unseen patrol boat below frame.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: charging crowd of Korean villagers in the upper-center of the frame, midground, moves toward lower dock and patrol boat; route toward the lower dock in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 선착장 위쪽에서 내려오는 길 (Occupied by villagers running toward the patrol boat) — Its downward axis runs from the upper background toward the lower center of frame; used as Descending route that organizes the crowd into depth and directs the charge toward camera; 선착장 하단 (The villagers are rapidly closing on it) — The approach side lies below the charging crowd and toward the camera; used as Lower spatial destination positioned close to the camera side of the action.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe main pier extends seaward from the village-side approach and terminates at an outer pier end surrounded by the adjacent sea. Its main deck provides a continuous route between the village and the outer end, while a vessel berth edge directly contacts the water along the boat-access side. At the landward end, the approach continues into the small island village. The fishing-gear warehouse and Banwol Restaurant face this pier-to-village route, with the warehouse doorway opening onto the landing-side escape path and the restaurant entrance addressing the same general frontage. Small village houses occupy the inland area around the approach, with woodland interwoven among and behind them. This arrangement allows views inward from the outer pier toward the village, outward from the village frontage toward the berth and sea, and lateral coverage between the warehouse, restaurant frontage, and main pier deck.\n- pier end: seaward terminal end of the island pier\n- pier: main island pier deck\n- berth: water-contacting vessel berth edge\n- water: sea adjoining the island pier\n- approach: village-to-pier approach\n- houses: small island village houses around the pier approach\n- woodland: woodland surrounding the small island village\n- warehouse: fishing-gear warehouse beside the pier landing\n- door: pier-facing fishing-gear warehouse doorway\n- restaurant: Banwol Restaurant at the village-side pier frontage\n- entrance: pier-visible entrance of Banwol Restaurant\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S86sh5__bgfirst_bg.png", "asset_id": "147e483b-b195-45db-b11d-910ed02f58f8", "input_asset_ids": [ "f069c325-d8f7-4ac1-b071-cab1f52171d8" ] }, "S86sh5": { "input_fingerprint": "b039f1cf8179d1c0", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장 위쪽에서 수리영 일행을 향해 몸을 굽히고 한 발을 뻗은 채 맹렬하게 뛰어내려오는 수많은 한국인 마을 사람들의 mid-action 전경.\n\nLOCATION (lock): The sloping village approach descending onto the island pier. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same island dock, village slope, harbor edge, and daylight. Exclude the earlier hostile stationary reception and show the villagers charging downhill toward the patrol boat.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain stained with goat blood as they run for the patrol boat. Jihu lacks her necklace, and Inwoo retains his backpack, cast, and revolver.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장 위쪽에서 수리영 일행을 향해 몸을 굽히고 한 발을 뻗은 채 맹렬하게 뛰어내려오는 수많은 한국인 마을 사람들의 mid-action 전경.\n\nLOCATION (lock): The sloping village approach descending onto the island pier. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain stained with goat blood as they run for the patrol boat. Jihu lacks her necklace, and Inwoo retains his backpack, cast, and revolver.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장 위쪽에서 수리영 일행을 향해 몸을 굽히고 한 발을 뻗은 채 맹렬하게 뛰어내려오는 수많은 한국인 마을 사람들의 mid-action 전경.\n\nLOCATION (lock): The sloping village approach descending onto the island pier. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain stained with goat blood as they run for the patrol boat. Jihu lacks her necklace, and Inwoo retains his backpack, cast, and revolver.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장 위쪽에서 수리영 일행을 향해 몸을 굽히고 한 발을 뻗은 채 맹렬하게 뛰어내려오는 수많은 한국인 마을 사람들의 mid-action 전경.\n\nLOCATION (lock): The sloping village approach descending onto the island pier. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu, Inwoo, and Minsook remain stained with goat blood as they run for the patrol boat. Jihu lacks her necklace, and Inwoo retains his backpack, cast, and revolver.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S86sh5__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S86sh5__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S86sh5__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "A": 7, "B": 4, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 전경 샷과 마을 사람들의 돌진 액션을 완벽히 구현했으며, 기준 인물(고산들)의 외양과 의상도 정확하게 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "달려오는 액션과 구도는 적절하나, 반드시 포함되어야 할 기준 인물(고산들)의 외양과 의상이 전혀 반영되지 않았습니다." }, { "label": "C", "score": 3, "verdict_ko": "등장이 금지된 인물(인우 등)이 프레임에 포함되었으며, 일행을 향해 달려오는 설정(방향성)과도 모순되는 치명적 위반이 있습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S86sh5__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S86sh5__bgfirst_bg.png", "bg_asset_id": "147e483b-b195-45db-b11d-910ed02f58f8", "bg_record_key": "S86sh5::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S85sh5" } }, "S86sh12": { "input_fingerprint": "0b34e6c6b37117f3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장 시멘트 바닥에 처박힌 채 피를 흘리며 널브러진 조 형사의 참혹한 시체.\n\nLOCATION (lock): The concrete pier between the village approach and the docked patrol boat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Detective Jo's corpse remains sprawled flat on the pier's concrete surface where it landed, with his neck broken, his head twisted out of alignment with his torso, and his limbs splayed motionless around him. The blood-covered fugitives continue toward the patrol boat.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'바닥에 처박혀 널브러진' 시체의 자세를 잘 연출했으며, 레퍼런스의 배경 구조물과 인물의 의상(모자 포함)을 가장 완벽하게 유지했습니다." }, { "label": "B", "score": 3, "verdict_ko": "시체가 참혹하게 처박힌 모습이 아니라 정자세로 반듯하게 누워 있어 제시문의 핵심 연출 의도를 완전히 무시했습니다." }, { "label": "C", "score": 4, "verdict_ko": "시체의 자세는 제시문에 부합하나, 선착장의 고정 구조물(볼라드) 형태가 레퍼런스와 다르고 인물의 모자가 누락되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S86sh5_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S86sh5" }, "lane_policy": "share_plan_prev_bgonly" }, "S86sh14": { "input_fingerprint": "6d9000978d63bb4c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 충격에 튀어나올 듯 커진 눈으로 굳어버린 신 형사의 경악한 얼굴.\n\nLOCATION (lock): Beside the patrol boat boarding point on the island pier. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dock, charging villagers, patrol boat, and daylight. Exclude the fallen detective from the close frame and isolate the younger detective's shocked reaction.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Detective Jo remains dead on the dock behind Shin, while Jihu, Inwoo, and Minsook board the patrol boat still coated in goat blood.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 신 형사 (30대 한국인 남성, 또렷한 얼굴, 활동적인 인상, 단정한 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 4, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "클로즈업 프레임을 통해 쓰러진 형사를 배제하고 신 형사의 경악한 얼굴을 단독으로 포착하라는 샷 지시를 완벽하게 이행했습니다." }, { "label": "A", "score": 4, "verdict_ko": "쓰러진 형사를 프레임에서 배제하라는 명시적인 지시를 어기고 배경에 그대로 노출시켰습니다." }, { "label": "B", "score": 3, "verdict_ko": "쓰러진 형사를 화면에 포함시켰을 뿐만 아니라, 이전 샷과 다르게 위치와 자세까지 임의로 변경하여 연속성을 훼손했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S86sh12_sel.png" }, { "label": "CHARACTER REFERENCE — 신 형사: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S86sh12" } }, "S87sh26::bgfirst_bg": { "input_fingerprint": "b395a7335252ac0e", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 다가오는 (짐승의 모습으로 변한 고산들) 앞에 서서 두 팔을 양옆으로 활짝 벌린 채 굳건하게 버티고 선 강민숙(홍은미)의 전신.\n\nLOCATION (lock): The open passage between the bow and stern on a moving police patrol boat.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the stern, the camera looks straight down the passage from behind 김지후(강수리영)’s position while keeping her outside the lower frame. 강민숙(홍은미) is held full-length in the center with both arms spread across the passage, while 고산들 (짐승 변이 상태) approaches in the deeper center along the same axis.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강민숙(홍은미) in the middle-center of the frame, midground; 고산들 (짐승 변이 상태) in the upper-center of the frame, background; stern passage separating them in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 선수와 선미를 연결하는 통로 (Occupied by the confrontation) — The passage recedes centrally from the stern toward 고산들 (짐승 변이 상태); used as Creates a compressed corridor between the protector and the approaching threat; 선미 난간 (Adjacent to 강민숙(홍은미)) — Its inner side runs along the lateral edge of the passage; used as Marks the lateral escape edge that will motivate the following move; 바다 (Visible beyond the vessel); used as Provides spatial depth beyond the rail without pulling focus from the standoff.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 다가오는 (짐승의 모습으로 변한 고산들) 앞에 서서 두 팔을 양옆으로 활짝 벌린 채 굳건하게 버티고 선 강민숙(홍은미)의 전신.\n\nLOCATION (lock): The open passage between the bow and stern on a moving police patrol boat.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level near the stern, the camera looks straight down the passage from behind 김지후(강수리영)’s position while keeping her outside the lower frame. 강민숙(홍은미) is held full-length in the center with both arms spread across the passage, while 고산들 (짐승 변이 상태) approaches in the deeper center along the same axis.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강민숙(홍은미) in the middle-center of the frame, midground; 고산들 (짐승 변이 상태) in the upper-center of the frame, background; stern passage separating them in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 선수와 선미를 연결하는 통로 (Occupied by the confrontation) — The passage recedes centrally from the stern toward 고산들 (짐승 변이 상태); used as Creates a compressed corridor between the protector and the approaching threat; 선미 난간 (Adjacent to 강민숙(홍은미)) — Its inner side runs along the lateral edge of the passage; used as Marks the lateral escape edge that will motivate the following move; 바다 (Visible beyond the vessel); used as Provides spatial depth beyond the rail without pulling focus from the standoff.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S87sh26__bgfirst_bg.png", "asset_id": "d18adfe1-73f2-4c88-9e49-f44812fe8dd9", "input_asset_ids": [ "66838c25-a59d-4ed7-bb90-8bc85ee25d3e", "dd2f6f40-02e4-4376-bb4e-eb1bd03ed128", "f01a72dd-1c7a-4752-9e6a-c272d65ec1c5" ] }, "S87sh26": { "input_fingerprint": "106898bb511cebe2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 다가오는 (짐승의 모습으로 변한 고산들) 앞에 서서 두 팔을 양옆으로 활짝 벌린 채 굳건하게 버티고 선 강민숙(홍은미)의 전신.\n\nLOCATION (lock): The open passage between the bow and stern on a moving police patrol boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains goat-blood-covered as she shields Jihu from Sandeul. Jihu is still necklace-less, while Inwoo retains his backpack, cast, and revolver until Sandeul knocks the gun away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (짐승 변이 상태) (한국인 남성형 괴물, 짧은 짙은 머리, 날카로운 얼굴, 짐승 같은 얼굴 윤곽); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 다가오는 (짐승의 모습으로 변한 고산들) 앞에 서서 두 팔을 양옆으로 활짝 벌린 채 굳건하게 버티고 선 강민숙(홍은미)의 전신.\n\nLOCATION (lock): The open passage between the bow and stern on a moving police patrol boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains goat-blood-covered as she shields Jihu from Sandeul. Jihu is still necklace-less, while Inwoo retains his backpack, cast, and revolver until Sandeul knocks the gun away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (짐승 변이 상태) (한국인 남성형 괴물, 짧은 짙은 머리, 날카로운 얼굴, 짐승 같은 얼굴 윤곽); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 다가오는 (짐승의 모습으로 변한 고산들) 앞에 서서 두 팔을 양옆으로 활짝 벌린 채 굳건하게 버티고 선 강민숙(홍은미)의 전신.\n\nLOCATION (lock): The open passage between the bow and stern on a moving police patrol boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook remains goat-blood-covered as she shields Jihu from Sandeul. Jihu is still necklace-less, while Inwoo retains his backpack, cast, and revolver until Sandeul knocks the gun away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (짐승 변이 상태) (한국인 남성형 괴물, 짧은 짙은 머리, 날카로운 얼굴, 짐승 같은 얼굴 윤곽); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S87sh26__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S87sh26.png" }, { "label": "CHARACTER REFERENCE — 고산들 (짐승 변이 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L45B04.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_police_patrol_boat_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들 (짐승 변이 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "괴물 앞을 가로막는 동작의 의도(상대방을 향해 서서 방어)를 정확히 연출했으며, 지정된 위치의 구조와 캐릭터 디자인을 충실히 반영했습니다." }, { "label": "B", "score": 3, "verdict_ko": "위치 레퍼런스에 없는 초록색 플라스틱 상자를 덱에 임의로 추가한 하드 위반이 있으며, 괴물을 등지고 서 있어 '앞을 가로막는' 연출 의도와 어긋나고 괴물에게 청바지가 입혀졌습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "괴물 앞을 가로막는 동작의 의도(상대방을 향해 서서 방어)를 정확히 연출했으며, 지정된 위치의 구조와 캐릭터 디자인을 충실히 반영했습니다." }, { "label": "B", "score": 3, "verdict_ko": "위치 레퍼런스에 없는 초록색 플라스틱 상자를 덱에 임의로 추가한 하드 위반이 있으며, 괴물을 등지고 서 있어 '앞을 가로막는' 연출 의도와 어긋나고 괴물에게 청바지가 입혀졌습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물들의 외형과 배경 디테일이 레퍼런스와 잘 일치하며, 두 팔을 벌리고 굳건히 선 전신 구도를 성공적으로 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "갑판 바닥의 해치 테두리가 인물의 부츠와 하나로 융합되어 물리적으로 불가능한 형태를 렌더링한 치명적인 오류가 있습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "인물들의 외형과 배경 디테일이 레퍼런스와 잘 일치하며, 두 팔을 벌리고 굳건히 선 전신 구도를 성공적으로 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "갑판 바닥의 해치 테두리가 인물의 부츠와 하나로 융합되어 물리적으로 불가능한 형태를 렌더링한 치명적인 오류가 있습니다." } ] }, "combined": { "totals": { "A": 10, "B": 10 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 10, "B": 10 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "괴물 앞을 가로막는 동작의 의도(상대방을 향해 서서 방어)를 정확히 연출했으며, 지정된 위치의 구조와 캐릭터 디자인을 충실히 반영했습니다." }, { "label": "B", "score": 3, "verdict_ko": "위치 레퍼런스에 없는 초록색 플라스틱 상자를 덱에 임의로 추가한 하드 위반이 있으며, 괴물을 등지고 서 있어 '앞을 가로막는' 연출 의도와 어긋나고 괴물에게 청바지가 입혀졌습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L45B04.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_police_patrol_boat_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들 (짐승 변이 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S87sh26__bgfirst_bg.png", "bg_asset_id": "d18adfe1-73f2-4c88-9e49-f44812fe8dd9", "bg_record_key": "S87sh26::bgfirst_bg", "chain_winner": true, "authority": "plate", "seed_attached": true }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S87sh34": { "input_fingerprint": "a48321835f6efec7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 서로 껴안은 상태로 난간을 넘어 허공에 비스듬히 기울어진 채 떠 있는 강민숙(홍은미)과 (짐승의 모습으로 변한 고산들)의 실루엣.\n\nLOCATION (lock): At the patrol boat’s side rail above the open sea. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same wet patrol-boat deck, rail position, churning sea, and daylight. Exclude the mother's blocking stance and show her locked with the transformed man as both tip over the rail.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook and Sandeul remain locked together as they go over the rail; both carry their existing blood and injuries into the fall.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (짐승 변이 상태) (한국인 남성형 괴물, 짧은 짙은 머리, 날카로운 얼굴, 짐승 같은 얼굴 윤곽); 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 4, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요청된 실루엣 연출과 난간 너머 허공에 비스듬히 떠 있는 껴안은 자세를 프롬프트에 맞게 가장 훌륭하게 구현했습니다." }, { "label": "C", "score": 4, "verdict_ko": "명시된 난간 구조물이 누락되었고 실루엣 효과가 부족하며, 공중에 떠 있는 자세의 물리적 설득력이 떨어집니다." }, { "label": "A", "score": 3, "verdict_ko": "여성의 두 발이 수직 벽면에 평평하게 붙어 있는 물리적으로 불가능한 자세를 묘사하여 심각한 위반에 해당합니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S87sh26_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들 (짐승 변이 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S87sh26" }, "lane_policy": "ab_select_bypass:prev" }, "S87sh41": { "input_fingerprint": "b788c2b9acc30027", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 난간에 기대어 엄마가 빠진 바다를 응시하며 눈물을 쏟아내는 김지후(강수리영)의 일그러진 옆얼굴.\n\nLOCATION (lock): The aft rail of the departing police patrol boat, overlooking its foaming wake. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same wet deck, railings, wake, and open sea under daylight. Exclude the transformed attacker and show the daughter leaning over the rail toward the water.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Minsook and Sandeul are submerged and no longer visible, leaving only foam at the fall point. Jihu remains goat-blood-stained and necklace-less; Inwoo retains his cast and backpack, but the revolver has been knocked away.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 8, "A": 5, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "샷 텍스트가 요구한 우는 옆얼굴의 감정을 잘 담아냈으며, 캐릭터 레퍼런스의 지정 복장(청바지와 재킷) 및 피 묻은 상태를 정확하게 재현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "구도와 감정 표현은 좋으나, 지정된 복장(청바지) 대신 이전 샷의 다른 인물이 입었던 붉은 치마를 입고 있어 복장 일관성에서 감점되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "프레임 우측에 샷 텍스트에 언급되지 않은 인물(깁스를 한 팔)이 등장하여 엑스트라 추가 금지 규정을 위반한 하드 바이올레이션입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S87sh34_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S87sh34" } }, "S88sh4::bgfirst_bg": { "input_fingerprint": "d4fae2a8e42e936c", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 선착장 위에 서 있는 강영란(흰 머리 여자)과 그 뒤로 빼곡히 모인 한국인 마을 사람들의 위협적인 모습.\n\nLOCATION (lock): The patrol boat’s open deck facing the opposite shore of the same island.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From standing eye level on the forward deck, the camera continues its dolly toward the landing point across the remaining water. 강영란(흰 머리 여자) holds the lower center as the foremost figure, while the densely packed 한국인 마을 사람들 rise behind her across the width of the landing area.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 선착장 (Densely occupied by villagers); used as Supports the hostile group as a dense horizontal mass behind 강영란(흰 머리 여자); 남아 있는 바다 (Visible between vessel and landing point); used as Maintains the physical separation between the patrol boat and the gathering; 순찰정 선수 (Approaching the landing point) — Its forward-facing section points toward the landing point; used as Provides a low foreground reference for the camera’s forward-deck position.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 선착장 위에 서 있는 강영란(흰 머리 여자)과 그 뒤로 빼곡히 모인 한국인 마을 사람들의 위협적인 모습.\n\nLOCATION (lock): The patrol boat’s open deck facing the opposite shore of the same island.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From standing eye level on the forward deck, the camera continues its dolly toward the landing point across the remaining water. 강영란(흰 머리 여자) holds the lower center as the foremost figure, while the densely packed 한국인 마을 사람들 rise behind her across the width of the landing area.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 선착장 (Densely occupied by villagers); used as Supports the hostile group as a dense horizontal mass behind 강영란(흰 머리 여자); 남아 있는 바다 (Visible between vessel and landing point); used as Maintains the physical separation between the patrol boat and the gathering; 순찰정 선수 (Approaching the landing point) — Its forward-facing section points toward the landing point; used as Provides a low foreground reference for the camera’s forward-deck position.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S88sh4__bgfirst_bg.png", "asset_id": "daa2285d-ea19-4074-aae2-2dcfc078d728", "input_asset_ids": [ "858d851c-3add-4ee7-8032-55cf572b1b40", "baf4016d-66f3-4b8d-bb0e-aedb66c32376" ] }, "S88sh4": { "input_fingerprint": "f4cb85407c119c93", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장 위에 서 있는 강영란(흰 머리 여자)과 그 뒤로 빼곡히 모인 한국인 마을 사람들의 위협적인 모습.\n\nLOCATION (lock): The patrol boat’s open deck facing the opposite shore of the same island. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul has reappeared alive on Geumwoldo after the fall, while Minsook remains missing. Jihu and Inwoo remain on the patrol boat, blood-stained, and Jihu still lacks her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장 위에 서 있는 강영란(흰 머리 여자)과 그 뒤로 빼곡히 모인 한국인 마을 사람들의 위협적인 모습.\n\nLOCATION (lock): The patrol boat’s open deck facing the opposite shore of the same island. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul has reappeared alive on Geumwoldo after the fall, while Minsook remains missing. Jihu and Inwoo remain on the patrol boat, blood-stained, and Jihu still lacks her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장 위에 서 있는 강영란(흰 머리 여자)과 그 뒤로 빼곡히 모인 한국인 마을 사람들의 위협적인 모습.\n\nLOCATION (lock): The patrol boat’s open deck facing the opposite shore of the same island. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul has reappeared alive on Geumwoldo after the fall, while Minsook remains missing. Jihu and Inwoo remain on the patrol boat, blood-stained, and Jihu still lacks her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S88sh4__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S88sh4.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L45B03.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "레퍼런스의 구도를 피하여 갑판 위에서 선착장과 위협적인 마을 사람들을 지시대로 잘 묘사했습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스의 조타실 내부 프레이밍을 그대로 복사하여 갑판 위라는 공간 지시와 구도 모방 금지 조건을 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "레퍼런스의 구도를 피하여 갑판 위에서 선착장과 위협적인 마을 사람들을 지시대로 잘 묘사했습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스의 조타실 내부 프레이밍을 그대로 복사하여 갑판 위라는 공간 지시와 구도 모방 금지 조건을 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 '야외 갑판' 위치를 잘 반영하였으며, 레퍼런스의 실내 구도를 복사하지 않고 마을 사람들의 위협적인 모습을 적절히 연출했습니다." }, { "label": "A", "score": 3, "verdict_ko": "레퍼런스 이미지의 실내 구도(창문과 계기판)를 그대로 복사하여 '야외 갑판'이라는 장소 지시 및 구도 모방 금지 규칙을 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '야외 갑판' 위치를 잘 반영하였으며, 레퍼런스의 실내 구도를 복사하지 않고 마을 사람들의 위협적인 모습을 적절히 연출했습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스 이미지의 실내 구도(창문과 계기판)를 그대로 복사하여 '야외 갑판'이라는 장소 지시 및 구도 모방 금지 규칙을 위반했습니다." } ] }, "combined": { "totals": { "A": 15, "B": 6 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 15, "B": 6 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "레퍼런스의 구도를 피하여 갑판 위에서 선착장과 위협적인 마을 사람들을 지시대로 잘 묘사했습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스의 조타실 내부 프레이밍을 그대로 복사하여 갑판 위라는 공간 지시와 구도 모방 금지 조건을 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L45B03.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S88sh4__bgfirst_bg.png", "bg_asset_id": "daa2285d-ea19-4074-aae2-2dcfc078d728", "bg_record_key": "S88sh4::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S88sh8": { "input_fingerprint": "ab4680549964af02", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 정인우의 머리를 향해 서늘한 총구를 바짝 들이민 채 고정된 구창모의 뻗은 팔.\n\nLOCATION (lock): On the patrol boat deck just outside the wheelhouse, within close range of the helm. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same boat deck, railings, cabin geometry, and approaching-island daylight. Exclude the earlier transformed attacker and show the local officer aiming a gun at the injured man.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Changmo keeps his handgun trained on Inwoo while steering the patrol boat back toward Geumwoldo. Jihu and Inwoo remain blood-stained, and Jihu remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 5, "B": 4, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "머리를 향해 팔을 뻗은 자세는 취했으나, 정작 손에 총이 없고 두 인물 간 거리가 멀어 '바짝 들이민' 텍스트 묘사를 온전히 구현하지 못했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물 간 거리는 가깝지만, 총을 겨누는 대신 어깨에 손을 올린 전혀 다른 행동을 묘사하여 핵심 동작을 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "하드 위반: 화면 우측 가장자리에 소품 레퍼런스 이미지의 회색 실루엣 손과 총이 그대로 유출되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S87sh41_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 38구경 권총: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S87sh41" } }, "S89sh1": { "input_fingerprint": "70ae97615a8b9a66", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 갯바위에 꼿꼿하게 선 채 순찰정을 향해 시선을 고정하고 만족스럽게 입꼬리를 올린 강영란(흰 머리 여자)의 얼굴.\n\nLOCATION (lock): A rocky landing on the island’s opposite shore, facing the approaching patrol boat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul remains alive beside the waiting villagers, while Minsook is still missing after the fall. Jihu's bone necklace remains under Youngran's control.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 5 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 '얼굴' 위주의 프레이밍에 가장 가깝게 연출되었으며, 인물 제한 및 복장 유지 규칙을 잘 준수했습니다." }, { "label": "B", "score": 3, "verdict_ko": "지문에 명시되지 않은 다수의 인물을 무단으로 추가하여 등장인물 제한 규칙을 심각하게 위반했습니다." }, { "label": "C", "score": 5, "verdict_ko": "불필요한 인물은 없으나, 요구된 '얼굴' 위주의 샷이 아닌 전신에 가까운 넓은 샷으로 연출되어 프레이밍에서 감점되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S88sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S88sh4" } }, "S90sh4::bgfirst_bg": { "input_fingerprint": "b1799371b3932c30", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 텅 빈 식당 한가운데 우두커니 서서 멍한 눈빛으로 허공을 응시하는 김지후(강수리영)의 전면.\n\nLOCATION (lock): The empty center of the restaurant hall at sunset, filled with soft reddish light from outside.\n\nTIME OF DAY (lock): sunset.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level in the middle of the empty dining hall, the camera holds directly in front of 김지후(강수리영) after completing the preceding move around her. She is centered in a medium frame with slack shoulders and planted feet, while empty space around her makes the unreturned stare feel isolated rather than portrait-like.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 김지후(강수리영) isolated in the dining hall in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 텅 빈 식당 홀 (Empty); used as The surrounding vacancy isolates 김지후(강수리영) and confirms that her search found no one; 주방 입구 (No person visible within) — The opening faces into the hall behind 김지후(강수리영); used as Retained behind her as one of the searched spaces; 방 입구 (No person visible within) — The room-facing opening is visible from the dining hall; used as Forms a second vacant depth layer associated with her search.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 텅 빈 식당 한가운데 우두커니 서서 멍한 눈빛으로 허공을 응시하는 김지후(강수리영)의 전면.\n\nLOCATION (lock): The empty center of the restaurant hall at sunset, filled with soft reddish light from outside.\n\nTIME OF DAY (lock): sunset.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level in the middle of the empty dining hall, the camera holds directly in front of 김지후(강수리영) after completing the preceding move around her. She is centered in a medium frame with slack shoulders and planted feet, while empty space around her makes the unreturned stare feel isolated rather than portrait-like.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 김지후(강수리영) isolated in the dining hall in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 텅 빈 식당 홀 (Empty); used as The surrounding vacancy isolates 김지후(강수리영) and confirms that her search found no one; 주방 입구 (No person visible within) — The opening faces into the hall behind 김지후(강수리영); used as Retained behind her as one of the searched spaces; 방 입구 (No person visible within) — The room-facing opening is visible from the dining hall; used as Forms a second vacant depth layer associated with her search.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S90sh4__bgfirst_bg.png", "asset_id": "89669719-11f1-4dbf-b701-25ab43eae81c", "input_asset_ids": [ "e1be9ccf-6c0a-417e-8889-8eb2a6150f46", "26ceb88e-2cc4-4e9d-97db-40b1d645b9ea" ] }, "S90sh4": { "input_fingerprint": "d22cb5c940bb035e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 식당 한가운데 우두커니 서서 멍한 눈빛으로 허공을 응시하는 김지후(강수리영)의 전면.\n\nLOCATION (lock): The empty center of the restaurant hall at sunset, filled with soft reddish light from outside. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): In the vision, Jihu remains without her bone necklace and searches the empty restaurant for Minsook.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 식당 한가운데 우두커니 서서 멍한 눈빛으로 허공을 응시하는 김지후(강수리영)의 전면.\n\nLOCATION (lock): The empty center of the restaurant hall at sunset, filled with soft reddish light from outside. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): In the vision, Jihu remains without her bone necklace and searches the empty restaurant for Minsook.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 식당 한가운데 우두커니 서서 멍한 눈빛으로 허공을 응시하는 김지후(강수리영)의 전면.\n\nLOCATION (lock): The empty center of the restaurant hall at sunset, filled with soft reddish light from outside. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): In the vision, Jihu remains without her bone necklace and searches the empty restaurant for Minsook.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S90sh4__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S90sh4.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B04.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "요구된 '멍한 눈빛'을 명확히 보여줄 수 있는 적절한 프레이밍이며, 목걸이가 없는 상태와 식당의 배경 디테일을 매우 완벽하게 재현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "배경과 인물 배치는 지시사항을 잘 따랐으나, 프레임이 다소 넓어 텍스트가 요구한 구체적인 표정 연기(멍한 눈빛)를 명확히 확인하기 어렵습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "요구된 '멍한 눈빛'을 명확히 보여줄 수 있는 적절한 프레이밍이며, 목걸이가 없는 상태와 식당의 배경 디테일을 매우 완벽하게 재현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "배경과 인물 배치는 지시사항을 잘 따랐으나, 프레임이 다소 넓어 텍스트가 요구한 구체적인 표정 연기(멍한 눈빛)를 명확히 확인하기 어렵습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 식당의 위치(벽면 메뉴판 등 세부 요소 포함)와 시간대, 캐릭터의 의상(뼈 목걸이 제거 상태) 및 멍한 표정을 완벽하게 재현했습니다." }, { "label": "B", "score": 7, "verdict_ko": "캐릭터와 의상, 조명은 훌륭하지만, 레퍼런스에 있던 벽면 메뉴판이 누락되는 등 배경 디테일의 정확도가 A보다 떨어집니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "지정된 식당의 위치(벽면 메뉴판 등 세부 요소 포함)와 시간대, 캐릭터의 의상(뼈 목걸이 제거 상태) 및 멍한 표정을 완벽하게 재현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "캐릭터와 의상, 조명은 훌륭하지만, 레퍼런스에 있던 벽면 메뉴판이 누락되는 등 배경 디테일의 정확도가 A보다 떨어집니다." } ] }, "combined": { "totals": { "A": 14, "B": 19 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 19 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "요구된 '멍한 눈빛'을 명확히 보여줄 수 있는 적절한 프레이밍이며, 목걸이가 없는 상태와 식당의 배경 디테일을 매우 완벽하게 재현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "배경과 인물 배치는 지시사항을 잘 따랐으나, 프레임이 다소 넓어 텍스트가 요구한 구체적인 표정 연기(멍한 눈빛)를 명확히 확인하기 어렵습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L20B04.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S90sh4__bgfirst_bg.png", "bg_asset_id": "89669719-11f1-4dbf-b701-25ab43eae81c", "bg_record_key": "S90sh4::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S91sh1::bgfirst_bg": { "input_fingerprint": "b7f102efd43ea1cc", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 노을이 지는 식당 뒤뜰에 서서 불안한 눈빛으로 한쪽을 돌아보는 김지후(강수리영)의 전신.\n\nLOCATION (lock): The restaurant’s rear yard in the fading light of sunset.\n\nTIME OF DAY (lock): sunset.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level and full-body distance, the camera tracks along a shallow arc as 김지후(강수리영) turns toward one side of the backyard. She remains near the center while her shifting torso and searching head create movement against the otherwise empty surrounding space.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 식당 뒤뜰 (No other person visible); used as Provides open negative space around her full figure, allowing each turn to read as an unsuccessful search.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 노을이 지는 식당 뒤뜰에 서서 불안한 눈빛으로 한쪽을 돌아보는 김지후(강수리영)의 전신.\n\nLOCATION (lock): The restaurant’s rear yard in the fading light of sunset.\n\nTIME OF DAY (lock): sunset.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At standing eye level and full-body distance, the camera tracks along a shallow arc as 김지후(강수리영) turns toward one side of the backyard. She remains near the center while her shifting torso and searching head create movement against the otherwise empty surrounding space.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 식당 뒤뜰 (No other person visible); used as Provides open negative space around her full figure, allowing each turn to read as an unsuccessful search.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S91sh1__bgfirst_bg.png", "asset_id": "fa2fef38-efe0-49e2-84c8-9cf8ecaf9046", "input_asset_ids": [ "52437600-957c-44f2-97fd-8f7f14392f4d", "87b15f17-cae5-428d-896d-70c058ca2a5f" ] }, "S91sh1": { "input_fingerprint": "a9e261c441251d51", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 노을이 지는 식당 뒤뜰에 서서 불안한 눈빛으로 한쪽을 돌아보는 김지후(강수리영)의 전신.\n\nLOCATION (lock): The restaurant’s rear yard in the fading light of sunset. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): In the vision, Jihu remains without her necklace as she continues searching the rear yard for Minsook.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 노을이 지는 식당 뒤뜰에 서서 불안한 눈빛으로 한쪽을 돌아보는 김지후(강수리영)의 전신.\n\nLOCATION (lock): The restaurant’s rear yard in the fading light of sunset. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): In the vision, Jihu remains without her necklace as she continues searching the rear yard for Minsook.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 노을이 지는 식당 뒤뜰에 서서 불안한 눈빛으로 한쪽을 돌아보는 김지후(강수리영)의 전신.\n\nLOCATION (lock): The restaurant’s rear yard in the fading light of sunset. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): In the vision, Jihu remains without her necklace as she continues searching the rear yard for Minsook.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S91sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S91sh1.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L31B02.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "요청된 앵글과 전신 샷을 정확히 구현했으며, 장소 레퍼런스의 디테일과 캐릭터의 외양 및 의상 스타일을 충실하게 유지했습니다." }, { "label": "B", "score": 6, "verdict_ko": "전신 샷과 표정은 잘 표현되었으나, 지정되지 않은 의상(맨발 포함)으로 변경되었고 배경의 소품(파란 상자 등)이 원본 레퍼런스와 다르게 변형되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "요청된 앵글과 전신 샷을 정확히 구현했으며, 장소 레퍼런스의 디테일과 캐릭터의 외양 및 의상 스타일을 충실하게 유지했습니다." }, { "label": "B", "score": 6, "verdict_ko": "전신 샷과 표정은 잘 표현되었으나, 지정되지 않은 의상(맨발 포함)으로 변경되었고 배경의 소품(파란 상자 등)이 원본 레퍼런스와 다르게 변형되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 전신 샷과 장소, 불안하게 돌아보는 포즈를 정확히 연출했으며 레퍼런스의 의상과도 일치하여 훌륭합니다." }, { "label": "A", "score": 6, "verdict_ko": "전신 샷과 표정 등 주요 지시사항은 잘 따랐으나, 캐릭터 레퍼런스와 전혀 다른 임의의 복장(맨발 등)을 착용하여 감점되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 전신 샷과 장소, 불안하게 돌아보는 포즈를 정확히 연출했으며 레퍼런스의 의상과도 일치하여 훌륭합니다." }, { "label": "B", "score": 6, "verdict_ko": "전신 샷과 표정 등 주요 지시사항은 잘 따랐으나, 캐릭터 레퍼런스와 전혀 다른 임의의 복장(맨발 등)을 착용하여 감점되었습니다." } ] }, "combined": { "totals": { "A": 16, "B": 12 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 16, "B": 12 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "요청된 앵글과 전신 샷을 정확히 구현했으며, 장소 레퍼런스의 디테일과 캐릭터의 외양 및 의상 스타일을 충실하게 유지했습니다." }, { "label": "B", "score": 6, "verdict_ko": "전신 샷과 표정은 잘 표현되었으나, 지정되지 않은 의상(맨발 포함)으로 변경되었고 배경의 소품(파란 상자 등)이 원본 레퍼런스와 다르게 변형되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L31B02.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S91sh1__bgfirst_bg.png", "bg_asset_id": "fa2fef38-efe0-49e2-84c8-9cf8ecaf9046", "bg_record_key": "S91sh1::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S92sh3::bgfirst_bg": { "input_fingerprint": "67a04b074de3c736", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 선착장 돌바닥에 앉아 고요히 바다를 쳐다보는 강민숙(홍은미)의 묵묵한 뒷모습.\n\nLOCATION (lock): The open area before the restaurant, looking down the village approach toward the pier.\n\nTIME OF DAY (lock): sunset.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The forward move settles at seated shoulder height directly behind 강민숙(홍은미), then becomes a static hold. Her quiet back occupies the lower center at medium distance, aligned symmetrically with the sea beyond so that no facial information interrupts the reveal.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강민숙(홍은미) seated from behind in the lower-center of the frame, midground; sea directly beyond 강민숙(홍은미) in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 선착장 돌바닥 (강민숙(홍은미) is seated on it); used as Supports her seated silhouette and occupies the lower portion without obscuring her body; 바다 (Visible beyond the landing area); used as Aligned directly beyond her back as the sole destination of her gaze; 선착장 가장자리 (Positioned ahead of 강민숙(홍은미)) — Its outer edge runs across the frame in front of the sea; used as Defines the threshold between the stone ground and the open view.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 선착장 돌바닥에 앉아 고요히 바다를 쳐다보는 강민숙(홍은미)의 묵묵한 뒷모습.\n\nLOCATION (lock): The open area before the restaurant, looking down the village approach toward the pier.\n\nTIME OF DAY (lock): sunset.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The forward move settles at seated shoulder height directly behind 강민숙(홍은미), then becomes a static hold. Her quiet back occupies the lower center at medium distance, aligned symmetrically with the sea beyond so that no facial information interrupts the reveal.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 강민숙(홍은미) seated from behind in the lower-center of the frame, midground; sea directly beyond 강민숙(홍은미) in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 선착장 돌바닥 (강민숙(홍은미) is seated on it); used as Supports her seated silhouette and occupies the lower portion without obscuring her body; 바다 (Visible beyond the landing area); used as Aligned directly beyond her back as the sole destination of her gaze; 선착장 가장자리 (Positioned ahead of 강민숙(홍은미)) — Its outer edge runs across the frame in front of the sea; used as Defines the threshold between the stone ground and the open view.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S92sh3__bgfirst_bg.png", "asset_id": "7d9c9a56-67b6-4807-92cd-1c76fb51cbc7", "input_asset_ids": [ "4b7ce915-9855-4ac7-ae3f-fb9400b6c262", "cf4ba1ec-34ab-40c7-9c9e-a21fe6299356" ] }, "S92sh3": { "input_fingerprint": "dd84af67dbb9eff7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 선착장 돌바닥에 앉아 고요히 바다를 쳐다보는 강민숙(홍은미)의 묵묵한 뒷모습.\n\nLOCATION (lock): The open area before the restaurant, looking down the village approach toward the pier. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Within Jihu's vision, the seated figure at the dock appears to be Minsook and cradles an unmoving person concealed in her arms.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 선착장 돌바닥에 앉아 고요히 바다를 쳐다보는 강민숙(홍은미)의 묵묵한 뒷모습.\n\nLOCATION (lock): The open area before the restaurant, looking down the village approach toward the pier. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Within Jihu's vision, the seated figure at the dock appears to be Minsook and cradles an unmoving person concealed in her arms.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 선착장 돌바닥에 앉아 고요히 바다를 쳐다보는 강민숙(홍은미)의 묵묵한 뒷모습.\n\nLOCATION (lock): The open area before the restaurant, looking down the village approach toward the pier. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Within Jihu's vision, the seated figure at the dock appears to be Minsook and cradles an unmoving person concealed in her arms.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙(홍은미) (40대 후반 한국인 여성, 갸름한 얼굴, 귀티 나는 이목구비, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S92sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S92sh3.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L23B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 장소의 공간감, 일몰 시간대, 돌바닥에 앉은 뒷모습 및 품에 무언가를 안고 있는 설정까지 프롬프트의 요구사항을 충실히 반영했습니다." }, { "label": "A", "score": 3, "verdict_ko": "다리의 해부학적 구조가 매우 부자연스러우며, 배경에 레퍼런스에 없는 단(턱)이 생성되어 공간 구조를 왜곡한 하드 위반입니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 장소의 공간감, 일몰 시간대, 돌바닥에 앉은 뒷모습 및 품에 무언가를 안고 있는 설정까지 프롬프트의 요구사항을 충실히 반영했습니다." }, { "label": "A", "score": 3, "verdict_ko": "다리의 해부학적 구조가 매우 부자연스러우며, 배경에 레퍼런스에 없는 단(턱)이 생성되어 공간 구조를 왜곡한 하드 위반입니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 식당 앞의 위치와 일몰 시간대를 훌륭하게 재현했으며, 안고 있는 인물을 품에 가려지도록 연출한 점이 지시사항과 잘 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "안고 있는 인물의 머리가 그대로 노출되어 '품에 감춰진' 상태를 위반했으며, 배경의 바닥에 원본에 없는 단차와 구조물이 추가되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 식당 앞의 위치와 일몰 시간대를 훌륭하게 재현했으며, 안고 있는 인물을 품에 가려지도록 연출한 점이 지시사항과 잘 일치합니다." }, { "label": "A", "score": 4, "verdict_ko": "안고 있는 인물의 머리가 그대로 노출되어 '품에 감춰진' 상태를 위반했으며, 배경의 바닥에 원본에 없는 단차와 구조물이 추가되었습니다." } ] }, "combined": { "totals": { "A": 7, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 7, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 장소의 공간감, 일몰 시간대, 돌바닥에 앉은 뒷모습 및 품에 무언가를 안고 있는 설정까지 프롬프트의 요구사항을 충실히 반영했습니다." }, { "label": "A", "score": 3, "verdict_ko": "다리의 해부학적 구조가 매우 부자연스러우며, 배경에 레퍼런스에 없는 단(턱)이 생성되어 공간 구조를 왜곡한 하드 위반입니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L23B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙(홍은미): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S92sh3__bgfirst_bg.png", "bg_asset_id": "7d9c9a56-67b6-4807-92cd-1c76fb51cbc7", "bg_record_key": "S92sh3::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S93sh4": { "input_fingerprint": "95e41a53c419733b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 고개를 뒤로 돌린 채 섬뜩한 미소를 지으며 김지후(강수리영)와 시선을 맞추고 있는 강영란(흰 머리 여자)의 정면 얼굴.\n\nLOCATION (lock): At the far end of the island pier overlooking the sunset sea. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same sunset dock, reddish sky, stone paving, and golden sea. Exclude the earlier corpse-investigation crowd and show the elder seated at the dock end turning toward the approaching woman.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The seated figure is revealed as Youngran, still holding an unmoving, waterlogged body in her arms.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 4, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '정면 얼굴'을 담기 위해 이전 샷과 다르게 카메라 구도를 성공적으로 변경했으며, 인물의 표정과 안고 있는 시체도 정확히 묘사했습니다." }, { "label": "B", "score": 4, "verdict_ko": "이전 샷의 넓은 카메라 구도를 그대로 답습하여 프레이밍을 변경하라는 지시를 어겼으며, 얼굴 또한 완전히 정면을 향하지 않습니다." }, { "label": "C", "score": 3, "verdict_ko": "레퍼런스의 구도를 복사해 '정면 얼굴' 샷을 구현하지 못하는 치명적인 오류를 범했으며, 얼굴이 측면에 더 가깝게 묘사되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S92sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S92sh3" } }, "S93sh8": { "input_fingerprint": "c7029869de71e0b1", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): sunset.\n\nSHOT TEXT (authoritative, Korean): 물에 흠뻑 젖은 채 생기 없이 늘어진 강민숙(홍은미)의 시체 클로즈업.\n\nLOCATION (lock): The stone pier edge beside the golden sunset water. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): In the vision, Minsook's soaked corpse hangs limp in Youngran's seated embrace, with her upper body slumped against Youngran's shoulder, her pale head dropping into view, and her arms and legs hanging without support.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 5, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "생기 없는 시체의 클로즈업을 잘 담아냈으며 지시된 어깨에 기댄 자세를 정확히 구현해 가장 우수합니다." }, { "label": "C", "score": 5, "verdict_ko": "시체의 표현과 자세는 적절하나, 지문에 명시되지 않은 산 사람(영란)의 턱과 목이 프레임에 노출되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "시체가 어깨에 늘어지지 않고 목이 잡힌 형태이며, 산 사람의 얼굴이 노출되어 지문과 어긋납니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S93sh4_sel.png" } ], "critique_skipped": true, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S93sh4" }, "lane_policy": "share_plan_prev_bgonly" }, "S94sh2::bgfirst_bg": { "input_fingerprint": "809cf4fc1d832af1", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 굵은 쇠사슬에 몸 전체가 단단히 결박된 채 바닥에 쓰러져 있는 김지후(강수리영)의 전신.\n\nLOCATION (lock): A rock-floored holding chamber deep inside the natural cave, dimly lit in daytime.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At floor level beside 김지후(강수리영)'s head, the camera finishes retreating along her body axis until her chained full body and its relation to the cave floor are readable in one low wide frame. Her head remains nearest the lens while the chain-bound limbs recede through the center, and the composition opens toward the opposite side in preparation for the curving track away.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: cave floor (beneath 김지후(강수리영)); used as Low plane supporting the full-body composition and guiding the retreating body axis; thick iron chains (fastened tightly around her body); used as Visible restraint pattern across the body; opposite cave wall (across the cave); used as Distant spatial boundary toward which the next camera move will travel.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 굵은 쇠사슬에 몸 전체가 단단히 결박된 채 바닥에 쓰러져 있는 김지후(강수리영)의 전신.\n\nLOCATION (lock): A rock-floored holding chamber deep inside the natural cave, dimly lit in daytime.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At floor level beside 김지후(강수리영)'s head, the camera finishes retreating along her body axis until her chained full body and its relation to the cave floor are readable in one low wide frame. Her head remains nearest the lens while the chain-bound limbs recede through the center, and the composition opens toward the opposite side in preparation for the curving track away.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: cave floor (beneath 김지후(강수리영)); used as Low plane supporting the full-body composition and guiding the retreating body axis; thick iron chains (fastened tightly around her body); used as Visible restraint pattern across the body; opposite cave wall (across the cave); used as Distant spatial boundary toward which the next camera move will travel.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S94sh2__bgfirst_bg.png", "asset_id": "c3d4e894-1200-45b0-bbc6-7e001143c333", "input_asset_ids": [ "02195201-7fc9-4d25-ba97-056c02fc6910", "3d2d81f3-41d9-4b12-98d1-8c3b51938aed" ] }, "S94sh2": { "input_fingerprint": "58050ef4056af8c8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 굵은 쇠사슬에 몸 전체가 단단히 결박된 채 바닥에 쓰러져 있는 김지후(강수리영)의 전신.\n\nLOCATION (lock): A rock-floored holding chamber deep inside the natural cave, dimly lit in daytime. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is collapsed on the cave floor with her entire body tightly secured by heavy chains, her torso resting on the ground and her bound arms and legs unable to extend.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu is collapsed on the cave floor with her entire body tightly secured by heavy chains, her torso resting on the ground and her bound arms and legs unable to extend, and remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 굵은 쇠사슬에 몸 전체가 단단히 결박된 채 바닥에 쓰러져 있는 김지후(강수리영)의 전신.\n\nLOCATION (lock): A rock-floored holding chamber deep inside the natural cave, dimly lit in daytime. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is collapsed on the cave floor with her entire body tightly secured by heavy chains, her torso resting on the ground and her bound arms and legs unable to extend.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu is collapsed on the cave floor with her entire body tightly secured by heavy chains, her torso resting on the ground and her bound arms and legs unable to extend, and remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 굵은 쇠사슬에 몸 전체가 단단히 결박된 채 바닥에 쓰러져 있는 김지후(강수리영)의 전신.\n\nLOCATION (lock): A rock-floored holding chamber deep inside the natural cave, dimly lit in daytime. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is collapsed on the cave floor with her entire body tightly secured by heavy chains, her torso resting on the ground and her bound arms and legs unable to extend.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu is collapsed on the cave floor with her entire body tightly secured by heavy chains, her torso resting on the ground and her bound arms and legs unable to extend, and remains without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S94sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S94sh2.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B03.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "전신 샷의 구도와 바닥에 완전히 쓰러진 자세를 명확히 구현했으며, 쇠사슬 소품의 재질과 동굴 배경이 레퍼런스와 매우 높은 일치도를 보입니다." }, { "label": "B", "score": 6, "verdict_ko": "전신이 프레임에 들어왔으나 머리가 바닥이 아닌 벽면 바위에 기대어 있고, 쇠사슬이 레퍼런스와 달리 심하게 녹슨 형태로 묘사되어 아쉽습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "전신 샷의 구도와 바닥에 완전히 쓰러진 자세를 명확히 구현했으며, 쇠사슬 소품의 재질과 동굴 배경이 레퍼런스와 매우 높은 일치도를 보입니다." }, { "label": "B", "score": 6, "verdict_ko": "전신이 프레임에 들어왔으나 머리가 바닥이 아닌 벽면 바위에 기대어 있고, 쇠사슬이 레퍼런스와 달리 심하게 녹슨 형태로 묘사되어 아쉽습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 '전신' 프레이밍을 여유 있는 구도로 정확히 구현했으며, 로케이션의 벽면 고정 장치와 사슬을 활용해 결박 상태를 자연스럽게 연출했습니다." }, { "label": "A", "score": 6, "verdict_ko": "인물의 전신이 프레임에 다소 꽉 차게 담겼으며, 몸을 감싼 사슬이 배경의 벽면 고정 장치와 연결되지 않아 공간과의 상호작용이 부족합니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 '전신' 프레이밍을 여유 있는 구도로 정확히 구현했으며, 로케이션의 벽면 고정 장치와 사슬을 활용해 결박 상태를 자연스럽게 연출했습니다." }, { "label": "B", "score": 6, "verdict_ko": "인물의 전신이 프레임에 다소 꽉 차게 담겼으며, 몸을 감싼 사슬이 배경의 벽면 고정 장치와 연결되지 않아 공간과의 상호작용이 부족합니다." } ] }, "combined": { "totals": { "A": 16, "B": 12 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 16, "B": 12 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "전신 샷의 구도와 바닥에 완전히 쓰러진 자세를 명확히 구현했으며, 쇠사슬 소품의 재질과 동굴 배경이 레퍼런스와 매우 높은 일치도를 보입니다." }, { "label": "B", "score": 6, "verdict_ko": "전신이 프레임에 들어왔으나 머리가 바닥이 아닌 벽면 바위에 기대어 있고, 쇠사슬이 레퍼런스와 달리 심하게 녹슨 형태로 묘사되어 아쉽습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B03.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S94sh2__bgfirst_bg.png", "bg_asset_id": "c3d4e894-1200-45b0-bbc6-7e001143c333", "bg_record_key": "S94sh2::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S94sh3": { "input_fingerprint": "b2c0007c0f6460e8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 반대쪽 벽에 쇠사슬로 묶인 채 피투성이가 된 얼굴로 김지후(강수리영)를 애타게 바라보는 정인우의 상체.\n\nLOCATION (lock): The opposite wall of the same cave holding chamber, fitted with heavy chains. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rough cave walls, daylight filtering inward, heavy chains, and damp floor. Exclude the woman's isolated prone framing and add the injured man chained to the opposite wall.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): He is restrained against the opposite cave wall by heavy chains, with his battered upper body held close to the wall and his arms confined at his sides.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo remains chained to the opposite wall with his face scratched and bloodied; Jihu remains chained across from him.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 4 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "벽에 밀착된 배치와 양팔이 고정된 지정 자세를 정확히 구현했으며, 요구된 상체 중심의 샷을 무난하게 살렸습니다." }, { "label": "B", "score": 5, "verdict_ko": "상체 프레이밍은 적절하나, 벽에 밀착되어야 한다는 지시와 달리 터널 중앙에 서 있는 배치 오류가 있습니다." }, { "label": "C", "score": 4, "verdict_ko": "정인우의 팔이 벌어져 고정 자세 지시에 어긋나며, 이전 샷에서 바닥에 누워있던 인물이 서 있는 모습으로 등장해 연속성을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S94sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S94sh2" } }, "groupbg::해안_절벽": { "input_fingerprint": "258fd1ff36624e96", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "해안_절벽", "tags": [ "S95sh1", "S95sh10", "S95sh15" ] }, "context_sig": "6442b56cc174ccbd" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: The exposed edge of a high coastal cliff above the blue sea in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 해안 절벽: 바다로 바로 떨어지는 수직의 높은 낭떠러지 위. (특징: 바다가 내려다보이는 높은 시야; 거친 암벽 바닥; 발아래 펼쳐진 푸른 바다)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 해안 절벽 - 낮\n- 붉은색 무녀복 차림의 영란과 황 씨가 절벽에 서서 바다를 바라본다.\n\nTIME OF DAY (lock): day, clear sky.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: The exposed edge of a high coastal cliff above the blue sea in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월도 해안 절벽: 바다로 바로 떨어지는 수직의 높은 낭떠러지 위. (특징: 바다가 내려다보이는 높은 시야; 거친 암벽 바닥; 발아래 펼쳐진 푸른 바다)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실외. 금월도 해안 절벽 - 낮\n- 붉은색 무녀복 차림의 영란과 황 씨가 절벽에 서서 바다를 바라본다.\n\nTIME OF DAY (lock): day, clear sky.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해안_절벽_9d2258.png", "asset_id": "c216ac4d-d8c0-4179-9f4a-fa3c564bfaa8", "input_asset_ids": [ "0d8fbc53-5607-4f6f-b79e-cf63bed82708" ], "origin_tag": "S95sh1", "place_text": "The exposed edge of a high coastal cliff above the blue sea in daylight.", "origin_inputs": { "place_text": "The exposed edge of a high coastal cliff above the blue sea in daylight.", "time_of_day_en": "day, clear sky", "conti_asset_id": "0d8fbc53-5607-4f6f-b79e-cf63bed82708" } }, "S95sh1::bgfirst_bg": { "input_fingerprint": "13c62cffbd7128fe", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 푸른 하늘과 바다를 배경으로 절벽 끝에 우뚝 서서 붉은 치마를 흩날리는 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): The exposed edge of a high coastal cliff above the blue sea in daylight.\n\nTIME OF DAY (lock): day, clear sky.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From standing eye height inland of the cliff, a static wide frame places 강영란(흰 머리 여자)'s full back near the center against the open sky and sea, preserving ample space around her and beyond the edge. Her planted figure forms a narrow crimson vertical in the landscape, while the frame's side remains open for the later approach and forward arc.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground, looks toward sea beyond the cliff; sky and sea beyond the cliff in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: blue sky (clear and blue); used as Broad upper and distant field silhouetting the solitary figure; blue sea (rolling beneath the sky); used as Deep background beyond the cliff edge; cliff edge (overlooking the sea); used as Hard spatial boundary beneath 강영란(흰 머리 여자)'s planted feet.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 푸른 하늘과 바다를 배경으로 절벽 끝에 우뚝 서서 붉은 치마를 흩날리는 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): The exposed edge of a high coastal cliff above the blue sea in daylight.\n\nTIME OF DAY (lock): day, clear sky.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From standing eye height inland of the cliff, a static wide frame places 강영란(흰 머리 여자)'s full back near the center against the open sky and sea, preserving ample space around her and beyond the edge. Her planted figure forms a narrow crimson vertical in the landscape, while the frame's side remains open for the later approach and forward arc.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강영란(흰 머리 여자) in the middle-center of the frame, midground, looks toward sea beyond the cliff; sky and sea beyond the cliff in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: blue sky (clear and blue); used as Broad upper and distant field silhouetting the solitary figure; blue sea (rolling beneath the sky); used as Deep background beyond the cliff edge; cliff edge (overlooking the sea); used as Hard spatial boundary beneath 강영란(흰 머리 여자)'s planted feet.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh1__bgfirst_bg.png", "asset_id": "5860251c-d378-4bb8-9d91-666665557b63", "input_asset_ids": [ "0d8fbc53-5607-4f6f-b79e-cf63bed82708", "c216ac4d-d8c0-4179-9f4a-fa3c564bfaa8" ] }, "S95sh1": { "input_fingerprint": "f08ec211c0641bbd", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, clear sky.\n\nSHOT TEXT (authoritative, Korean): 푸른 하늘과 바다를 배경으로 절벽 끝에 우뚝 서서 붉은 치마를 흩날리는 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): The exposed edge of a high coastal cliff above the blue sea in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe. Sandeul, who survived the sea, remains beside her before the attack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, clear sky.\n\nSHOT TEXT (authoritative, Korean): 푸른 하늘과 바다를 배경으로 절벽 끝에 우뚝 서서 붉은 치마를 흩날리는 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): The exposed edge of a high coastal cliff above the blue sea in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe. Sandeul, who survived the sea, remains beside her before the attack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, clear sky.\n\nSHOT TEXT (authoritative, Korean): 푸른 하늘과 바다를 배경으로 절벽 끝에 우뚝 서서 붉은 치마를 흩날리는 강영란(흰 머리 여자)의 꼿꼿한 뒷모습.\n\nLOCATION (lock): The exposed edge of a high coastal cliff above the blue sea in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran continues wearing the red shaman's robe. Sandeul, who survived the sea, remains beside her before the attack.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S95sh1.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해안_절벽_9d2258.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 인물이 추가되어 인물 제한 규정을 위반한 치명적 오류가 있습니다." }, { "label": "B", "score": 7, "verdict_ko": "텍스트에 명시된 인물만 정확히 단독으로 배치했으며, 뒷모습 프레이밍과 흩날리는 레퍼런스 의상의 디테일을 훌륭히 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 인물이 추가되어 인물 제한 규정을 위반한 치명적 오류가 있습니다." }, { "label": "B", "score": 7, "verdict_ko": "텍스트에 명시된 인물만 정확히 단독으로 배치했으며, 뒷모습 프레이밍과 흩날리는 레퍼런스 의상의 디테일을 훌륭히 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "샷 텍스트의 지시대로 화면에 강영란 한 명만을 정확히 배치했으며, 붉은 치마가 흩날리는 뒷모습을 위치 배경에 맞춰 훌륭하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 텍스트가 인물의 등장을 결정한다는 규칙을 어기고 텍스트에 없는 인물을 화면에 추가하여 심각한 감점 요인이 되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "샷 텍스트의 지시대로 화면에 강영란 한 명만을 정확히 배치했으며, 붉은 치마가 흩날리는 뒷모습을 위치 배경에 맞춰 훌륭하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트가 인물의 등장을 결정한다는 규칙을 어기고 텍스트에 없는 인물을 화면에 추가하여 심각한 감점 요인이 되었습니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 인물이 추가되어 인물 제한 규정을 위반한 치명적 오류가 있습니다." }, { "label": "B", "score": 7, "verdict_ko": "텍스트에 명시된 인물만 정확히 단독으로 배치했으며, 뒷모습 프레이밍과 흩날리는 레퍼런스 의상의 디테일을 훌륭히 구현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해안_절벽_9d2258.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh1__bgfirst_bg.png", "bg_asset_id": "5860251c-d378-4bb8-9d91-666665557b63", "bg_record_key": "S95sh1::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "해안_절벽", "groupbg_asset_id": "c216ac4d-d8c0-4179-9f4a-fa3c564bfaa8" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S95sh10": { "input_fingerprint": "1061bbfa6d2c8137", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, clear sky.\n\nSHOT TEXT (authoritative, Korean): 날카로운 칼날이 고산들의 등 한가운데를 파고든 찰나, 칼자루를 꽉 쥔 강영란(흰 머리 여자)의 힘 들어간 뻗은 팔.\n\nLOCATION (lock): At the brink of the island’s high sea cliff. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cliff edge, blue sea, clear sky, and wind-blown red clothing. Exclude the calm seaward stance and show the elder stabbing the man from behind.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran's blade is embedded in Sandeul's back, creating a deep bleeding wound that must carry into all subsequent appearances.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 2 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "칼날이 남자의 등에 박힌 핵심 동작과 뻗은 팔, 두 인물의 의상 및 상태를 지시문과 레퍼런스에 맞게 정확히 구현함." }, { "label": "B", "score": 3, "verdict_ko": "칼이 남자의 등에 박혀 있지 않고 허공을 향해 있으며, 남자의 가죽 재킷 의상이 레퍼런스와 불일치함." }, { "label": "C", "score": 2, "verdict_ko": "칼이 남자의 등에 박혀 있지 않으며, 이전 샷 상태에 없던 모자가 여자에게 임의로 추가되는 오류가 있음." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 식칼·도축용 칼·의식용 칼: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S95sh1" } }, "S95sh15::bgfirst_bg": { "input_fingerprint": "dd6561f384c80a41", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 강한 발길질의 충격으로 절벽 밖 허공에 붕 뜬 채 뒤로 기우는 고산들의 몸.\n\nLOCATION (lock): Beyond the cliff edge over the rocky sea far below.\n\nTIME OF DAY (lock): day, clear sky.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height and parallel to the cliff edge, the tracking camera begins wide enough to hold 고산들's entire body as the kick carries him backward beyond solid ground. His airborne body tilts across the middle-right of frame, with the cliff edge retained below and open space visibly extending behind him so the direction and danger of the fall are unmistakable.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 고산들 in the middle-right of the frame, midground, moves toward open space beyond the cliff edge; open space beyond the cliff edge in the middle-right of the frame, background; cliff edge in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: cliff edge (beneath and behind his airborne body); used as Required boundary proving that 고산들 has left solid ground; space beyond the cliff (open beneath the falling path); used as Open depth into which the tracking move continues; sea below (beneath the cliff); used as Distant endpoint of the fall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is organized along an inland-to-seaward axis. The inland forest margin opens directly onto the cliff-top standing ground, allowing movement from the forest to the overlook. The standing ground extends seaward to a distinct high brink. Beyond that brink, the cliff drops directly to the sea. Rocks occupy the water at the foot of the drop, where they remain visible from above. The principal top-side camera area is the open ground between the forest margin and the brink, while the lower reef and waterline form the downward view beyond the edge.\n- cliff top: Seaward-facing cliff-top standing ground\n- cliff edge: High seaward cliff brink and drop\n- forest margin: Inland forest margin opening onto the cliff top\n- sea: Sea directly below the coastal cliff\n- rock outcrop: Rocks at the foot of the cliff\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 강한 발길질의 충격으로 절벽 밖 허공에 붕 뜬 채 뒤로 기우는 고산들의 몸.\n\nLOCATION (lock): Beyond the cliff edge over the rocky sea far below.\n\nTIME OF DAY (lock): day, clear sky.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height and parallel to the cliff edge, the tracking camera begins wide enough to hold 고산들's entire body as the kick carries him backward beyond solid ground. His airborne body tilts across the middle-right of frame, with the cliff edge retained below and open space visibly extending behind him so the direction and danger of the fall are unmistakable.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 고산들 in the middle-right of the frame, midground, moves toward open space beyond the cliff edge; open space beyond the cliff edge in the middle-right of the frame, background; cliff edge in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: cliff edge (beneath and behind his airborne body); used as Required boundary proving that 고산들 has left solid ground; space beyond the cliff (open beneath the falling path); used as Open depth into which the tracking move continues; sea below (beneath the cliff); used as Distant endpoint of the fall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is organized along an inland-to-seaward axis. The inland forest margin opens directly onto the cliff-top standing ground, allowing movement from the forest to the overlook. The standing ground extends seaward to a distinct high brink. Beyond that brink, the cliff drops directly to the sea. Rocks occupy the water at the foot of the drop, where they remain visible from above. The principal top-side camera area is the open ground between the forest margin and the brink, while the lower reef and waterline form the downward view beyond the edge.\n- cliff top: Seaward-facing cliff-top standing ground\n- cliff edge: High seaward cliff brink and drop\n- forest margin: Inland forest margin opening onto the cliff top\n- sea: Sea directly below the coastal cliff\n- rock outcrop: Rocks at the foot of the cliff\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh15__bgfirst_bg.png", "asset_id": "14b08a5a-8582-463c-b918-523bd85aff29", "input_asset_ids": [ "8663c499-d6ea-4313-a17c-d1c983b5fb39" ] }, "S95sh15": { "input_fingerprint": "b9ee554e78814b17", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, clear sky.\n\nSHOT TEXT (authoritative, Korean): 강한 발길질의 충격으로 절벽 밖 허공에 붕 뜬 채 뒤로 기우는 고산들의 몸.\n\nLOCATION (lock): Beyond the cliff edge over the rocky sea far below. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same sheer cliff, blue water below, sky, and strong coastal wind. Exclude the stabbing contact and show the man already kicked out over the drop.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul falls from the cliff with deep stab wounds in both his back and chest; the wounds and blood remain with him.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, clear sky.\n\nSHOT TEXT (authoritative, Korean): 강한 발길질의 충격으로 절벽 밖 허공에 붕 뜬 채 뒤로 기우는 고산들의 몸.\n\nLOCATION (lock): Beyond the cliff edge over the rocky sea far below. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul falls from the cliff with deep stab wounds in both his back and chest; the wounds and blood remain with him.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, clear sky.\n\nSHOT TEXT (authoritative, Korean): 강한 발길질의 충격으로 절벽 밖 허공에 붕 뜬 채 뒤로 기우는 고산들의 몸.\n\nLOCATION (lock): Beyond the cliff edge over the rocky sea far below. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul falls from the cliff with deep stab wounds in both his back and chest; the wounds and blood remain with him.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day, clear sky.\n\nSHOT TEXT (authoritative, Korean): 강한 발길질의 충격으로 절벽 밖 허공에 붕 뜬 채 뒤로 기우는 고산들의 몸.\n\nLOCATION (lock): Beyond the cliff edge over the rocky sea far below. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul falls from the cliff with deep stab wounds in both his back and chest; the wounds and blood remain with him.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh15__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh15__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh15__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "C": 7, "A": 6, "B": 5 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "배경 레퍼런스와 더미의 포즈를 완벽히 일치시켰으며, 캐릭터의 이목구비, 헤어스타일, 상처 디테일의 재현도가 가장 높습니다." }, { "label": "A", "score": 6, "verdict_ko": "전반적인 연출과 재현도가 훌륭하지만, 손가락의 형태나 헤어스타일이 레퍼런스와 미세한 차이를 보입니다." }, { "label": "B", "score": 5, "verdict_ko": "배경과 공중 포즈는 잘 구현되었으나, 캐릭터의 헤어스타일이 레퍼런스 인물보다 지나치게 짧게 묘사되었습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh15__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S95sh15__bgfirst_bg.png", "bg_asset_id": "14b08a5a-8582-463c-b918-523bd85aff29", "bg_record_key": "S95sh15::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S95sh10" } }, "S96sh5": { "input_fingerprint": "6589fef8cb6fac9a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 강영란(흰 머리 여자)을 향해 고개를 치켜든 채 원망 섞인 눈빛을 쏘아보내는 김지후(강수리영)의 붉어진 얼굴.\n\nLOCATION (lock): The chained holding chamber inside the cave, illuminated by a carried lantern. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cave walls, chains, damp floor, and cool daylight. Exclude the earlier unconscious posture and show the chained woman glaring up at the elder.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains chained on the cave floor without her necklace, and Inwoo remains chained to the opposite wall with a battered face.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "얼굴 중심의 프레이밍과 랜턴을 든 인물을 향한 시선을 가장 잘 구현했으나(우선순위 2), 지후가 반대편 벽에 위치하는 오류(우선순위 3)가 있습니다." }, { "label": "A", "score": 5, "verdict_ko": "프레이밍이 상대적으로 넓어 얼굴에 온전히 집중되지 않으며, 랜턴이 바닥에 방치되어 들고 있다는 지침(우선순위 3)을 어겼습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍이 너무 넓어 텍스트의 의도를 크게 벗어났으며(우선순위 2), 지후가 금지된 목걸이를 착용하고 있습니다(우선순위 5)." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S94sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S94sh3" } }, "S96sh7": { "input_fingerprint": "1fb01eb2eb92f2a7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 두 눈을 커다랗게 뜬 채 딱딱하게 굳어버린 김지후(강수리영)의 경악한 얼굴.\n\nLOCATION (lock): On the floor of the lantern-lit cave chamber beside the captive’s chains. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rough stone, chain restraints, subdued daylight, and dark recesses. Exclude the elder from the immediate close frame and isolate the woman's shocked reaction.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu and Inwoo remain chained in the cave; Jihu's neck is still bare because Youngran retains the necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 8, "A": 6, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "이전 샷의 자세와 사슬 위치를 완벽히 유지하면서 지시문대로 인물의 경악한 얼굴을 클로즈업으로 훌륭하게 고립시켰습니다." }, { "label": "A", "score": 6, "verdict_ko": "얼굴 표정과 자세의 연속성은 좋으나, 노인을 배제하라는 지시에도 불구하고 화면 좌측 상단에 랜턴이 남아있어 시선을 다소 분산시킵니다." }, { "label": "B", "score": 3, "verdict_ko": "클로즈업 지시를 어기고 프레임이 넓어졌으며, 이전 샷에서 고정된 손과 사슬의 위치가 일치하지 않고 랜턴이 공중에 떠 있는 심각한 오류가 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S96sh5_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S96sh5" } }, "S97sh1": { "input_fingerprint": "a0033883e8b1c87e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 가슴팍에 하얀 사슴뿔 창끝이 깊게 꽂혀 밖으로 튀어나와 있는 그분(검은 연기 형체)의 기괴한 상체.\n\nLOCATION (lock): The corpse-strewn ground directly outside the old reed-field storehouse in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): In the historical flashback, the black entity remains impaled through the heart by the white antler-tipped spear and grips the Joseon village chief's neck.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 3, "B": 8, "C": 1 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "사슴뿔 창이 가슴에 꽂혀 있지 않고 허리에 차고 있으며, 화면 우측 하단에 카메라가 노출되는 치명적인 오류가 있습니다." }, { "label": "B", "score": 8, "verdict_ko": "가슴팍에 튀어나온 사슴뿔 창끝과 검은 연기에 휩싸인 상체의 모습을 제시된 프레이밍에 맞게 가장 잘 구현했습니다." }, { "label": "C", "score": 1, "verdict_ko": "요구된 상체 샷이 아닌 엎드린 인물을 묘사했으며, 다리가 여러 개로 얽힌 물리적으로 불가능한 신체 구조를 보여줍니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L03B01.png" } ], "critique_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } }, "S97sh5": { "input_fingerprint": "7e39c6ebc4045ac3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 두 눈이 기괴한 황금빛으로 번뜩이는 조선인 촌주의 얼굴 클로즈업.\n\nLOCATION (lock): Before the reed-field storehouse among the surviving villagers and fallen soldiers. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same reed-filled clearing, old wooden shed, daylight, and fallen bodies. Exclude the impaled smoke figure from the close frame and show the village leader's newly golden eyes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The entity has vanished, leaving the white antler-tipped spear behind, while the village chief's eyes remain transformed to gold.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 조선인 촌주 (50대 중반 조선인 남성, 각진 중년 얼굴, 단정한 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "C": 5, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "캐릭터의 얼굴과 회색 의상, 바닥에 남겨진 뿔창을 가장 정확히 구현했으나 레퍼런스의 머리띠가 누락되었습니다." }, { "label": "C", "score": 5, "verdict_ko": "머리띠와 의상은 일치하나 레퍼런스에 없는 수염이 추가되어 인물의 얼굴이 달라졌습니다." }, { "label": "A", "score": 3, "verdict_ko": "프롬프트가 명시적으로 배제하라고 지시한 이전 장면의 인물(검은 옷)을 촌주로 잘못 등장시키는 치명적 오류를 범했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S97sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 조선인 촌주: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S97sh1" } }, "S98sh2": { "input_fingerprint": "99e844d0bbcd0960", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 눈물이 고인 채 분노 가득한 눈빛으로 강영란(흰 머리 여자)을 쏘아보는 김지후(강수리영)의 일그러진 얼굴.\n\nLOCATION (lock): The captive’s chained position inside the dim natural-cave chamber. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rough cave, chain fixtures, damp floor, and cool daylight. Exclude the earlier collapsed posture and show the woman confronting the elder through tears.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains chained and necklace-less, while Inwoo remains chained to the opposite wall with his injured face.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "샷 텍스트에 명시된 두 인물(김지후, 강영란)만을 정확히 프레임에 담았으며, 지시된 눈물과 분노 섞인 표정 연출을 훌륭하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 텍스트에 언급되지 않은 제3의 인물(배경의 남성)을 프레임에 임의로 포함시켜 인물 노출 제한 규칙을 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "샷 텍스트에 없는 인물을 배경에 등장시켰을 뿐만 아니라, 인물의 손과 사슬의 접촉 방식이 기준 이미지의 상태를 벗어나 위반입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S96sh7_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S96sh7" } }, "S98sh9": { "input_fingerprint": "dbf98093e5a0a07c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두운 동굴 안에서 시니컬하게 미소 짓는 강영란(흰 머리 여자)과 그를 원망스레 쏘아보는 김지후(강수리영)의 팽팽한 대치 전경.\n\nLOCATION (lock): The central holding chamber inside the cave, between the lantern-bearing visitor and chained captives. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same stone walls, chained positions, dim recesses, and filtered daylight. Exclude the earlier rescue framing and stage the elder and captive woman in a tense face-to-face composition.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu and Inwoo remain chained in their established positions throughout the confrontation; Youngran still controls Jihu's missing necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 4, "A": 2 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "이전 샷에서 벽에 기대어 사슬을 쥔 지후의 고정된 자세를 정확히 유지했으며, 지시된 두 인물간의 팽팽한 대치 전경을 충실히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "지시된 두 인물만 잘 담아냈으나, 이전 샷에서 벽에 기대어 있던 지후가 동굴 중앙으로 이동하고 손의 결박 위치가 바뀌는 등 고정 자세 유지 지침을 어겼습니다." }, { "label": "A", "score": 2, "verdict_ko": "프레임에 없어야 할 제3의 인물(배경의 남자)이 등장한 심각한 위반이 있으며, 지후가 일어서 있어 이전 샷의 자세와 전혀 일치하지 않습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S98sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S98sh2" } }, "groupbg::해안_지구대_사무실": { "input_fingerprint": "f2a157feb66ad107", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "해안_지구대_사무실", "tags": [ "S99sh1", "S99sh4" ] }, "context_sig": "5ae83f4251daeb61" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: At a desk inside a small coastal police substation under daytime office lighting.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안 마을 경찰 지구대 내부: 시골 섬마을의 허름한 관할 파출소 실내. (특징: 낡은 책상과 의자; 서류철과 전화기; 벽면에 붙은 지도와 게시물)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실내. 해안마을 경찰 지구대 - 낮\n- 따르르릉- 전화 소리. 느긋하게 앉아 핸드폰 게임을 하는 창모.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: At a desk inside a small coastal police substation under daytime office lighting.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안 마을 경찰 지구대 내부: 시골 섬마을의 허름한 관할 파출소 실내. (특징: 낡은 책상과 의자; 서류철과 전화기; 벽면에 붙은 지도와 게시물)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 실내. 해안마을 경찰 지구대 - 낮\n- 따르르릉- 전화 소리. 느긋하게 앉아 핸드폰 게임을 하는 창모.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해안_지구대_사무실_ba5a20.png", "asset_id": "410f1c7a-f7da-4097-9402-a5ec7a454d04", "input_asset_ids": [ "59747ce5-67f8-40d4-94e0-099971d50a12" ], "origin_tag": "S99sh1", "place_text": "At a desk inside a small coastal police substation under daytime office lighting.", "origin_inputs": { "place_text": "At a desk inside a small coastal police substation under daytime office lighting.", "time_of_day_en": "day", "conti_asset_id": "59747ce5-67f8-40d4-94e0-099971d50a12" } }, "S99sh1::bgfirst_bg": { "input_fingerprint": "71de523819ccbad5", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 지구대 책상에 느긋하게 기댄 채 핸드폰 화면에 푹 빠져 있는 구창모의 상체.\n\nLOCATION (lock): At a desk inside a small coastal police substation under daytime office lighting.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold the flow's static seated-chest-height view from across the desk, framing 구창모 from the waist up at a slight frontal three-quarter angle as he reclines and looks down at his phone. Place the active phone low in the center of his torso area, with the desk edge crossing the lower frame and enough station context remaining behind him for the coming lateral pan.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 지구대 책상 (구창모가 기대어 있는 상태) — Its near edge faces the camera and runs across the lower frame; used as Lower-frame spatial divider between camera and 구창모; 핸드폰 화면 (게임이 실행 중) — The display face is angled upward toward 구창모 and partly toward the camera, showing the game he is playing; used as Primary point of attention held naturally in 구창모's hands.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 지구대 책상에 느긋하게 기댄 채 핸드폰 화면에 푹 빠져 있는 구창모의 상체.\n\nLOCATION (lock): At a desk inside a small coastal police substation under daytime office lighting.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold the flow's static seated-chest-height view from across the desk, framing 구창모 from the waist up at a slight frontal three-quarter angle as he reclines and looks down at his phone. Place the active phone low in the center of his torso area, with the desk edge crossing the lower frame and enough station context remaining behind him for the coming lateral pan.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 지구대 책상 (구창모가 기대어 있는 상태) — Its near edge faces the camera and runs across the lower frame; used as Lower-frame spatial divider between camera and 구창모; 핸드폰 화면 (게임이 실행 중) — The display face is angled upward toward 구창모 and partly toward the camera, showing the game he is playing; used as Primary point of attention held naturally in 구창모's hands.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S99sh1__bgfirst_bg.png", "asset_id": "02f74d99-6e53-4eb5-8361-1f3ae06648bb", "input_asset_ids": [ "59747ce5-67f8-40d4-94e0-099971d50a12", "410f1c7a-f7da-4097-9402-a5ec7a454d04" ] }, "S99sh1": { "input_fingerprint": "dc57afc60e15b3d8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 지구대 책상에 느긋하게 기댄 채 핸드폰 화면에 푹 빠져 있는 구창모의 상체.\n\nLOCATION (lock): At a desk inside a small coastal police substation under daytime office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 지구대 책상에 느긋하게 기댄 채 핸드폰 화면에 푹 빠져 있는 구창모의 상체.\n\nLOCATION (lock): At a desk inside a small coastal police substation under daytime office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 지구대 책상에 느긋하게 기댄 채 핸드폰 화면에 푹 빠져 있는 구창모의 상체.\n\nLOCATION (lock): At a desk inside a small coastal police substation under daytime office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S99sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S99sh1.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해안_지구대_사무실_ba5a20.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "책상에 기대어 핸드폰에 집중하는 모습을 잘 표현했으며, 소품의 방향(화면이 인물을 향함)도 규칙에 맞게 정확히 연출되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "인물이 핸드폰을 보고 있으나 화면(전면 카메라 노치 부분)이 렌즈를 향하고 있어 소품 방향 지침에 대한 치명적인 위반입니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "책상에 기대어 핸드폰에 집중하는 모습을 잘 표현했으며, 소품의 방향(화면이 인물을 향함)도 규칙에 맞게 정확히 연출되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "인물이 핸드폰을 보고 있으나 화면(전면 카메라 노치 부분)이 렌즈를 향하고 있어 소품 방향 지침에 대한 치명적인 위반입니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "새로운 카메라 앵글로 장소를 재현하며 책상에 기댄 인물의 포즈와 상체 프레이밍을 정확히 구현함." }, { "label": "B", "score": 3, "verdict_ko": "인물 연출은 무난하나, 장소 레퍼런스의 카메라 구도를 그대로 복사하지 말라는 지시를 명백히 위반함." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "새로운 카메라 앵글로 장소를 재현하며 책상에 기댄 인물의 포즈와 상체 프레이밍을 정확히 구현함." }, { "label": "A", "score": 3, "verdict_ko": "인물 연출은 무난하나, 장소 레퍼런스의 카메라 구도를 그대로 복사하지 말라는 지시를 명백히 위반함." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "책상에 기대어 핸드폰에 집중하는 모습을 잘 표현했으며, 소품의 방향(화면이 인물을 향함)도 규칙에 맞게 정확히 연출되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "인물이 핸드폰을 보고 있으나 화면(전면 카메라 노치 부분)이 렌즈를 향하고 있어 소품 방향 지침에 대한 치명적인 위반입니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해안_지구대_사무실_ba5a20.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S99sh1__bgfirst_bg.png", "bg_asset_id": "02f74d99-6e53-4eb5-8361-1f3ae06648bb", "bg_record_key": "S99sh1::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "해안_지구대_사무실", "groupbg_asset_id": "410f1c7a-f7da-4097-9402-a5ec7a454d04" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S99sh4": { "input_fingerprint": "4405f551fd6da2db", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수화기를 귀에 밀착한 채 입술을 꾹 다문 구창모의 심각한 얼굴.\n\nLOCATION (lock): The duty desk and landline area inside the coastal police substation. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same police desk, office furnishings, daylight, and relaxed station atmosphere. Exclude the mobile game action and show the officer taking the landline call seriously.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 4, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "이전 샷의 공간 연속성을 가장 잘 유지하고 지시된 심각한 표정을 정확히 연출했으나, 책상 위 수화기가 중복 생성된 오류가 있습니다." }, { "label": "B", "score": 4, "verdict_ko": "공간 연속성은 어느 정도 유지했으나, 수화기를 든 오른손의 해부학적 형태가 심하게 훼손되는 치명적 위반이 발생했습니다." }, { "label": "C", "score": 3, "verdict_ko": "이전 샷의 위치(캐비닛 앞)를 무시하고 의자에 앉은 구도로 임의 변경했으며, 지시되지 않은 소품(수첩, 펜)을 추가해 연속성을 훼손했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S99sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S99sh1" } }, "S100sh3::bgfirst_bg": { "input_fingerprint": "b8bb31106d2249a4", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한국인 여자들 앞에서 양손을 허리에 얹고 턱을 치켜든 채 서 있는 구창모의 전면.\n\nLOCATION (lock): A mainland harbor parking and unloading area beside a dark passenger van.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Maintain the flow's static eye-level address position directly in front of 구창모, framing him from mid-thigh upward with the 한국인 여자 무리 distributed behind and to either side beyond him. His hands lock at his waist and his raised chin points toward the women near the camera line, while the parked black van remains a modest background scale reference rather than a foreground obstruction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 까만색 그랜드카니발 (선착장에 정차한 상태) — A side portion of the parked vehicle faces the camera behind the group; used as Background context for the group's arrival and a scale reference behind the women; 선착장 (인물들 아래와 뒤로 보임); used as Ground plane connecting 구창모 and the women in one address space.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한국인 여자들 앞에서 양손을 허리에 얹고 턱을 치켜든 채 서 있는 구창모의 전면.\n\nLOCATION (lock): A mainland harbor parking and unloading area beside a dark passenger van.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Maintain the flow's static eye-level address position directly in front of 구창모, framing him from mid-thigh upward with the 한국인 여자 무리 distributed behind and to either side beyond him. His hands lock at his waist and his raised chin points toward the women near the camera line, while the parked black van remains a modest background scale reference rather than a foreground obstruction.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 까만색 그랜드카니발 (선착장에 정차한 상태) — A side portion of the parked vehicle faces the camera behind the group; used as Background context for the group's arrival and a scale reference behind the women; 선착장 (인물들 아래와 뒤로 보임); used as Ground plane connecting 구창모 and the women in one address space.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh3__bgfirst_bg.png", "asset_id": "883be430-0482-405e-bc09-15bdc6722465", "input_asset_ids": [ "45b50c3a-cbc3-4985-bcd9-521df664c303", "a146bae0-100d-4928-b618-164368b367c6" ] }, "S100sh3": { "input_fingerprint": "4d724664aeb2c7ba", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한국인 여자들 앞에서 양손을 허리에 얹고 턱을 치켜든 채 서 있는 구창모의 전면.\n\nLOCATION (lock): A mainland harbor parking and unloading area beside a dark passenger van. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The five or six women remain gathered with Changmo at the dock for transport to Geumwoldo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한국인 여자들 앞에서 양손을 허리에 얹고 턱을 치켜든 채 서 있는 구창모의 전면.\n\nLOCATION (lock): A mainland harbor parking and unloading area beside a dark passenger van. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The five or six women remain gathered with Changmo at the dock for transport to Geumwoldo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 한국인 여자들 앞에서 양손을 허리에 얹고 턱을 치켜든 채 서 있는 구창모의 전면.\n\nLOCATION (lock): A mainland harbor parking and unloading area beside a dark passenger van. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The five or six women remain gathered with Changmo at the dock for transport to Geumwoldo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S100sh3.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_육지_항구_선착장_19d1cf.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 턱을 치켜들고 허리에 손을 얹은 구창모의 정면 구도를 훌륭하게 소화했으며, 레퍼런스의 부두 배경과 구조물을 정확하게 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물들의 자세와 소품은 무난하나, 레퍼런스에 없는 2층 건물을 부두 왼쪽에 임의로 생성하여 장소 일치 조건(Location Lock)을 크게 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 턱을 치켜들고 허리에 손을 얹은 구창모의 정면 구도를 훌륭하게 소화했으며, 레퍼런스의 부두 배경과 구조물을 정확하게 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물들의 자세와 소품은 무난하나, 레퍼런스에 없는 2층 건물을 부두 왼쪽에 임의로 생성하여 장소 일치 조건(Location Lock)을 크게 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 구도를 잘 따르며, 5명의 여성 인원수와 구창모의 크로스백 등 레퍼런스의 디테일을 충실히 반영했습니다." }, { "label": "B", "score": 5, "verdict_ko": "구창모의 전면 구도는 좋으나 크로스백이 누락되었고, 명시된 여성의 인원수(4명)가 부족하여 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 구도를 잘 따르며, 5명의 여성 인원수와 구창모의 크로스백 등 레퍼런스의 디테일을 충실히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "구창모의 전면 구도는 좋으나 크로스백이 누락되었고, 명시된 여성의 인원수(4명)가 부족하여 감점되었습니다." } ] }, "combined": { "totals": { "A": 14, "B": 12 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 14, "B": 12 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 턱을 치켜들고 허리에 손을 얹은 구창모의 정면 구도를 훌륭하게 소화했으며, 레퍼런스의 부두 배경과 구조물을 정확하게 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물들의 자세와 소품은 무난하나, 레퍼런스에 없는 2층 건물을 부두 왼쪽에 임의로 생성하여 장소 일치 조건(Location Lock)을 크게 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_육지_항구_선착장_19d1cf.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh3__bgfirst_bg.png", "bg_asset_id": "883be430-0482-405e-bc09-15bdc6722465", "bg_record_key": "S100sh3::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "육지_항구_선착장", "groupbg_asset_id": "a146bae0-100d-4928-b618-164368b367c6" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S100sh5::bgfirst_bg": { "input_fingerprint": "f8d4281c8778278f", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 낚시 어선 앞에 서서 구창모를 향해 한 손을 머리 위로 번쩍 들어 올린 채 서 있는 황봉업의 전신.\n\nLOCATION (lock): The boarding section of the mainland pier beside a waiting fishing boat.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Let the flow's pan settle at eye level near 구창모's group, using a sliver of 구창모's shoulder at the left foreground to motivate the long view across the dock. 황봉업 appears full-length in the middle-right before the fishing boat, one arm raised overhead and his body turned toward 구창모, with surrounding dock space preserving the intended distance.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 구창모 in the middle-left of the frame, foreground, looks toward 황봉업; 황봉업 in the middle-right of the frame, midground; 낚시 어선 in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 낚시 어선 (선착장에 있는 상태) — Its side and the area in front of it are presented across the dock behind 황봉업; used as Large background anchor immediately behind 황봉업 without exceeding natural scale; 선착장 (두 인물 사이로 길게 보임); used as Long spatial corridor connecting 구창모's foreground position to 황봉업.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is organized on a single landside-to-water axis. The landside harbor access area meets the inland entrance of the main coastal village harbor dock, with a fixed landing identification sign positioned at that transition. The dock extends outward from this access area toward its outer terminus. Its boat-facing edge directly adjoins the harbor water along the working length of the landing, and the water also continues around the dock terminus. Camera positions used by the scenes can therefore progress from the landside access area onto the main dock and farther toward the outer terminus, while views toward berthed vessels are oriented across the boat-facing edge toward the adjoining harbor water.\n- approach: Landside harbor access area\n- sign: Fixed harbor landing identification sign\n- dock: Main coastal village harbor dock\n- berthing edge: Boat-facing berthing edge of the main dock\n- terminus: Outer terminus of the harbor dock\n- water: Harbor water beside the landing\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 낚시 어선 앞에 서서 구창모를 향해 한 손을 머리 위로 번쩍 들어 올린 채 서 있는 황봉업의 전신.\n\nLOCATION (lock): The boarding section of the mainland pier beside a waiting fishing boat.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Let the flow's pan settle at eye level near 구창모's group, using a sliver of 구창모's shoulder at the left foreground to motivate the long view across the dock. 황봉업 appears full-length in the middle-right before the fishing boat, one arm raised overhead and his body turned toward 구창모, with surrounding dock space preserving the intended distance.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 구창모 in the middle-left of the frame, foreground, looks toward 황봉업; 황봉업 in the middle-right of the frame, midground; 낚시 어선 in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 낚시 어선 (선착장에 있는 상태) — Its side and the area in front of it are presented across the dock behind 황봉업; used as Large background anchor immediately behind 황봉업 without exceeding natural scale; 선착장 (두 인물 사이로 길게 보임); used as Long spatial corridor connecting 구창모's foreground position to 황봉업.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is organized on a single landside-to-water axis. The landside harbor access area meets the inland entrance of the main coastal village harbor dock, with a fixed landing identification sign positioned at that transition. The dock extends outward from this access area toward its outer terminus. Its boat-facing edge directly adjoins the harbor water along the working length of the landing, and the water also continues around the dock terminus. Camera positions used by the scenes can therefore progress from the landside access area onto the main dock and farther toward the outer terminus, while views toward berthed vessels are oriented across the boat-facing edge toward the adjoining harbor water.\n- approach: Landside harbor access area\n- sign: Fixed harbor landing identification sign\n- dock: Main coastal village harbor dock\n- berthing edge: Boat-facing berthing edge of the main dock\n- terminus: Outer terminus of the harbor dock\n- water: Harbor water beside the landing\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh5__bgfirst_bg.png", "asset_id": "0d1a466d-004d-44be-8062-7ebdf018ffd4", "input_asset_ids": [ "e767ade4-f2f4-4ceb-baa6-3e0d131436a4" ] }, "S100sh5": { "input_fingerprint": "83a57b0bcf442692", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낚시 어선 앞에 서서 구창모를 향해 한 손을 머리 위로 번쩍 들어 올린 채 서 있는 황봉업의 전신.\n\nLOCATION (lock): The boarding section of the mainland pier beside a waiting fishing boat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dock surface, fishing boat, harbor equipment, and daylight. Exclude the disguised policeman's earlier stance and show the large boat operator waving from beside the vessel.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낚시 어선 앞에 서서 구창모를 향해 한 손을 머리 위로 번쩍 들어 올린 채 서 있는 황봉업의 전신.\n\nLOCATION (lock): The boarding section of the mainland pier beside a waiting fishing boat. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낚시 어선 앞에 서서 구창모를 향해 한 손을 머리 위로 번쩍 들어 올린 채 서 있는 황봉업의 전신.\n\nLOCATION (lock): The boarding section of the mainland pier beside a waiting fishing boat. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 낚시 어선 앞에 서서 구창모를 향해 한 손을 머리 위로 번쩍 들어 올린 채 서 있는 황봉업의 전신.\n\nLOCATION (lock): The boarding section of the mainland pier beside a waiting fishing boat. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 황봉업 (50대 중반 한국인 남성, 넉넉한 인상, 둥근 큰 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh5__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh5__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh5__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "C": 7, "A": 4, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 인물 구성과 황봉업의 의상, 자세를 레퍼런스에 맞게 정확히 구현한 가장 우수한 결과물입니다." }, { "label": "A", "score": 4, "verdict_ko": "구도와 자세는 적절하나 황봉업의 의상이 레퍼런스와 일치하지 않고, 배에 금지된 텍스트가 삽입되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "황봉업의 의상은 맞췄으나, 샷 텍스트에 없는 인물들을 추가로 생성하는 치명적인 위반(Hard Violation)을 범했습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh5__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 황봉업: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh5__bgfirst_bg.png", "bg_asset_id": "0d1a466d-004d-44be-8062-7ebdf018ffd4", "bg_record_key": "S100sh5::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S100sh3" } }, "S101sh1::bgfirst_bg": { "input_fingerprint": "53d0520ab2c4acc7", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어선 위, 한국인 낚시꾼들과 짙은 화장의 한국인 여자들로 빼곡하게 만원인 배의 전경.\n\nLOCATION (lock): The crowded open deck of a fishing boat moored at the mainland harbor.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Begin the flow in a static eye-level wide view from several meters off the boat's side, containing the full passenger area and enough of the vessel to make its overcrowding immediately legible. Arrange the 한국인 낚시꾼들 and 한국인 여자들 in layered rows across the passenger area, with bodies overlapping naturally but faces and gestures remaining intermittently readable.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 낚시 어선 (승객들로 만원인 상태) — The boat's side faces the camera, exposing the full passenger area above it; used as Primary environmental container and compositional base for the crowded passengers; 인접한 선착장 (배 옆에 부분적으로 보임); used as Secondary spatial edge beside the boat, preparing the later track toward land.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어선 위, 한국인 낚시꾼들과 짙은 화장의 한국인 여자들로 빼곡하게 만원인 배의 전경.\n\nLOCATION (lock): The crowded open deck of a fishing boat moored at the mainland harbor.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Begin the flow in a static eye-level wide view from several meters off the boat's side, containing the full passenger area and enough of the vessel to make its overcrowding immediately legible. Arrange the 한국인 낚시꾼들 and 한국인 여자들 in layered rows across the passenger area, with bodies overlapping naturally but faces and gestures remaining intermittently readable.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 낚시 어선 (승객들로 만원인 상태) — The boat's side faces the camera, exposing the full passenger area above it; used as Primary environmental container and compositional base for the crowded passengers; 인접한 선착장 (배 옆에 부분적으로 보임); used as Secondary spatial edge beside the boat, preparing the later track toward land.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S101sh1__bgfirst_bg.png", "asset_id": "8e38d9b5-5f9c-43f1-a306-c82cd45910ec", "input_asset_ids": [ "c3c6ac60-f09e-465a-83b4-a2cebe328214", "4c5651e0-d76f-4acd-b49c-f201c1e4168c", "e6f2c697-eafa-4ffc-b339-d0a0588aecb4" ] }, "S101sh1": { "input_fingerprint": "3f949f7c282357f8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어선 위, 한국인 낚시꾼들과 짙은 화장의 한국인 여자들로 빼곡하게 만원인 배의 전경.\n\nLOCATION (lock): The crowded open deck of a fishing boat moored at the mainland harbor. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fishing boat remains crowded with the bar women and anglers being taken to Geumwoldo as festival victims.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 첫 항해 낚싯꾼#1 (한국인 성인 남성, 그을린 얼굴, 짧은 짙은 머리); 첫 항해 낚싯꾼#2 (한국인 성인 남성, 평범한 얼굴, 짧은 짙은 머리); 첫 항해 낚싯꾼#3 (한국인 성인 남성, 둥근 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어선 위, 한국인 낚시꾼들과 짙은 화장의 한국인 여자들로 빼곡하게 만원인 배의 전경.\n\nLOCATION (lock): The crowded open deck of a fishing boat moored at the mainland harbor. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fishing boat remains crowded with the bar women and anglers being taken to Geumwoldo as festival victims.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 첫 항해 낚싯꾼#1 (한국인 성인 남성, 그을린 얼굴, 짧은 짙은 머리); 첫 항해 낚싯꾼#2 (한국인 성인 남성, 평범한 얼굴, 짧은 짙은 머리); 첫 항해 낚싯꾼#3 (한국인 성인 남성, 둥근 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어선 위, 한국인 낚시꾼들과 짙은 화장의 한국인 여자들로 빼곡하게 만원인 배의 전경.\n\nLOCATION (lock): The crowded open deck of a fishing boat moored at the mainland harbor. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fishing boat remains crowded with the bar women and anglers being taken to Geumwoldo as festival victims.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 첫 항해 낚싯꾼#1 (한국인 성인 남성, 그을린 얼굴, 짧은 짙은 머리); 첫 항해 낚싯꾼#2 (한국인 성인 남성, 평범한 얼굴, 짧은 짙은 머리); 첫 항해 낚싯꾼#3 (한국인 성인 남성, 둥근 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S101sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S101sh1.png" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L16B02.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_fishing_boat_sel.png" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 구조물(해풍호)의 형태를 정확히 반영했으며, 짙은 화장의 여성들과 낚시꾼들이 만원인 배의 전경을 충실히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 선박의 형태를 무시하고 지붕에 인물을 배치하는 등 구조를 임의로 변형하여 우선순위 지침을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 구조물(해풍호)의 형태를 정확히 반영했으며, 짙은 화장의 여성들과 낚시꾼들이 만원인 배의 전경을 충실히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 선박의 형태를 무시하고 지붕에 인물을 배치하는 등 구조를 임의로 변형하여 우선순위 지침을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "구조물 레퍼런스(해풍호)의 외형을 정확히 반영했으며, 낚시꾼들과 짙은 화장의 여성들로 붐비는 어선의 전경을 지문대로 충실히 묘사했습니다." }, { "label": "B", "score": 4, "verdict_ko": "지문에 없는 강아지와 부두 노동자가 임의로 추가되었고, 조타실 지붕 위에 사람이 올라가 있는 비현실적인 배치가 포함되어 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "구조물 레퍼런스(해풍호)의 외형을 정확히 반영했으며, 낚시꾼들과 짙은 화장의 여성들로 붐비는 어선의 전경을 지문대로 충실히 묘사했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지문에 없는 강아지와 부두 노동자가 임의로 추가되었고, 조타실 지붕 위에 사람이 올라가 있는 비현실적인 배치가 포함되어 감점되었습니다." } ] }, "combined": { "totals": { "A": 8, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 8, "B": 15 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 구조물(해풍호)의 형태를 정확히 반영했으며, 짙은 화장의 여성들과 낚시꾼들이 만원인 배의 전경을 충실히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 선박의 형태를 무시하고 지붕에 인물을 배치하는 등 구조를 임의로 변형하여 우선순위 지침을 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L16B02.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_fishing_boat_sel.png" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S101sh1__bgfirst_bg.png", "bg_asset_id": "8e38d9b5-5f9c-43f1-a306-c82cd45910ec", "bg_record_key": "S101sh1::bgfirst_bg", "chain_winner": false, "authority": "plate", "seed_attached": true }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S101sh6": { "input_fingerprint": "029a85c6f64cfb3f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 선착장에 서서 뚱한 표정으로 짐을 든 채 배를 노려보는 구창모의 전신.\n\nLOCATION (lock): On the mainland pier alongside the packed fishing boat. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 3 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 전신(Full body) 샷의 프레이밍을 정확히 구현했으며, 짐을 들고 배를 노려보는 뚱한 표정도 잘 살려냈습니다." }, { "label": "A", "score": 5, "verdict_ko": "인물의 표정과 행동은 적절하나, 전신 샷 지시와 달리 발목 아래가 잘린 프레이밍으로 인해 감점되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "지시문에 없는 다수의 인물(배 위의 선원들)을 화면에 임의로 추가하여 심각한 위반(Hard violation)을 범했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S100sh5_sel.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S100sh5" } }, "S101sh15": { "input_fingerprint": "0d1b02c0544c56fb", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 눈을 동그랗게 뜬 채 급격한 호기심을 드러내는 구창모의 붉어진 얼굴.\n\nLOCATION (lock): The dockside beside the fishing boat’s boarding rail. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same crowded fishing boat, deck fittings, harbor daylight, and gathered passengers. Exclude the broad crowd emphasis and isolate the local officer's sudden curiosity near the boat.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 '붉어진 얼굴'과 표정에 집중한 적절한 타이트한 구도이며, 지정된 부두 위치와 기존 설정을 가장 잘 유지했습니다." }, { "label": "B", "score": 5, "verdict_ko": "구도가 넓어 텍스트가 강조한 얼굴 묘사의 집중도가 떨어지며, 이전 컷에 없던 나무 경사로가 임의로 추가되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "인물의 위치를 부두가 아닌 배 위로 완전히 이동시켜 장소 고정 및 연속성 지시를 심각하게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S101sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S101sh6" } }, "S102sh1": { "input_fingerprint": "e9bb6840e3c8d6a6", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning to evening.\n\nSHOT TEXT (authoritative, Korean): 어두운 동굴 안, 핏대가 선 얼굴로 굵은 쇠사슬을 있는 힘껏 당기는 자세로 정지한 정인우의 일그러진 상체.\n\nLOCATION (lock): The chained wall area inside the dark cave holding chamber. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same rough cave walls, heavy chains, damp floor, and cool daylight. Exclude the woman's collapsed pose from prominence and show the man straining against his restraints.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): He is seated heavily on the cave floor facing away from the anchoring point, with his torso pitched backward and both hands pulling a thick chain taut while his lower body remains confined.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Inwoo remains chained in the cave with his face battered, while Jihu remains chained opposite him without her necklace.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 4, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "레퍼런스의 복장(조끼와 셔츠)을 정확히 유지하고 핏대가 선 얼굴과 사슬을 당기는 자세를 훌륭히 구현했으나, 우측 가장자리에 지시문에 없는 신체 일부가 약간 노출됨." }, { "label": "A", "score": 4, "verdict_ko": "사슬을 당기는 인물의 표정과 근육의 긴장감은 잘 표현되었으나, 이전 샷에서 유지되어야 할 복장 지침을 무시하고 상의를 탈의한 상태로 렌더링됨." }, { "label": "C", "score": 3, "verdict_ko": "캐릭터의 복장이 레퍼런스와 전혀 일치하지 않으며, 샷 텍스트에 명시되지 않은 인물(여성)이 배경에 등장하여 등장인물 제한 지침을 명백히 위반함." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S98sh9_sel.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S98sh9" } }, "S102sh16::bgfirst_bg": { "input_fingerprint": "3e06d6067a415a62", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어창 안, 손발이 묶이고 입이 틀어막힌 채 공포에 질려 올려다보는 십수 명의 한국인 여자들의 창백한 얼굴들.\n\nLOCATION (lock): Inside a cramped fish hold beneath the fishing boat’s deck, exposed by an opened hatch.\n\nTIME OF DAY (lock): day transitioning to evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From inside the opened boat compartment, hold the camera just above the captives' face level at the completed upward tilt, placing the hatch opening across the upper frame and the clustered faces through the middle and lower frame. The nearest bound women are held in three-quarter angles while overlapping rows of upturned Korean faces convey the number of captives without losing the enclosure.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 어창 뚜껑과 입구 (open) — The hatch is open above the camera-facing interior, exposing the opening over the captives; used as Creates the upper boundary of the confined composition and establishes the women's upward gaze; 어창 내부 (occupied by restrained captives); used as Crowds the women together and limits visible depth behind their faces; 손발의 결박과 입을 막은 구속물 (secured around the captives); used as Makes the captivity legible along the lower and side edges of the group.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어창 안, 손발이 묶이고 입이 틀어막힌 채 공포에 질려 올려다보는 십수 명의 한국인 여자들의 창백한 얼굴들.\n\nLOCATION (lock): Inside a cramped fish hold beneath the fishing boat’s deck, exposed by an opened hatch.\n\nTIME OF DAY (lock): day transitioning to evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From inside the opened boat compartment, hold the camera just above the captives' face level at the completed upward tilt, placing the hatch opening across the upper frame and the clustered faces through the middle and lower frame. The nearest bound women are held in three-quarter angles while overlapping rows of upturned Korean faces convey the number of captives without losing the enclosure.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 어창 뚜껑과 입구 (open) — The hatch is open above the camera-facing interior, exposing the opening over the captives; used as Creates the upper boundary of the confined composition and establishes the women's upward gaze; 어창 내부 (occupied by restrained captives); used as Crowds the women together and limits visible depth behind their faces; 손발의 결박과 입을 막은 구속물 (secured around the captives); used as Makes the captivity legible along the lower and side edges of the group.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S102sh16__bgfirst_bg.png", "asset_id": "c42a39ef-8a1f-424d-99a3-7b65743b53af", "input_asset_ids": [ "a43ae568-214d-4eb4-90c0-ece388b925cb", "632fe42f-04cf-468e-bcea-baa7b045ecbd" ] }, "S102sh16": { "input_fingerprint": "5a5fd94d5152a5b2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning to evening.\n\nSHOT TEXT (authoritative, Korean): 어창 안, 손발이 묶이고 입이 틀어막힌 채 공포에 질려 올려다보는 십수 명의 한국인 여자들의 창백한 얼굴들.\n\nLOCATION (lock): Inside a cramped fish hold beneath the fishing boat’s deck, exposed by an opened hatch. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The women are packed together inside the fishing hold in cramped seated or curled positions, with every woman's hands and feet bound and each mouth gagged; their torsos face upward toward the opened hatch.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The additional women remain packed together inside the fishing hold in cramped seated or curled positions, with every woman's hands and feet bound and each mouth gagged; their torsos face upward toward the opened hatch.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning to evening.\n\nSHOT TEXT (authoritative, Korean): 어창 안, 손발이 묶이고 입이 틀어막힌 채 공포에 질려 올려다보는 십수 명의 한국인 여자들의 창백한 얼굴들.\n\nLOCATION (lock): Inside a cramped fish hold beneath the fishing boat’s deck, exposed by an opened hatch. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The women are packed together inside the fishing hold in cramped seated or curled positions, with every woman's hands and feet bound and each mouth gagged; their torsos face upward toward the opened hatch.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The additional women remain packed together inside the fishing hold in cramped seated or curled positions, with every woman's hands and feet bound and each mouth gagged; their torsos face upward toward the opened hatch.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning to evening.\n\nSHOT TEXT (authoritative, Korean): 어창 안, 손발이 묶이고 입이 틀어막힌 채 공포에 질려 올려다보는 십수 명의 한국인 여자들의 창백한 얼굴들.\n\nLOCATION (lock): Inside a cramped fish hold beneath the fishing boat’s deck, exposed by an opened hatch. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): The women are packed together inside the fishing hold in cramped seated or curled positions, with every woman's hands and feet bound and each mouth gagged; their torsos face upward toward the opened hatch.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The additional women remain packed together inside the fishing hold in cramped seated or curled positions, with every woman's hands and feet bound and each mouth gagged; their torsos face upward toward the opened hatch.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정인우 (20대 중반 한국인 남성, 훤칠하고 잘생긴 얼굴, 단정한 짙은색 짧은 머리); 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S102sh16__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S102sh16.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B05.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 어창 배경과 쇠사슬 소품을 정확히 포함하였으며, 공포에 질려 위를 올려다보는 창백한 얼굴들을 성공적으로 연출했습니다." }, { "label": "A", "score": 3, "verdict_ko": "전경에 위치한 인물 주변에 주인을 알 수 없는 발이 나타나는 해부학적 오류가 있으며, 필수 소품인 쇠사슬이 누락되었고 표정에서 공포감이 부족합니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 어창 배경과 쇠사슬 소품을 정확히 포함하였으며, 공포에 질려 위를 올려다보는 창백한 얼굴들을 성공적으로 연출했습니다." }, { "label": "A", "score": 3, "verdict_ko": "전경에 위치한 인물 주변에 주인을 알 수 없는 발이 나타나는 해부학적 오류가 있으며, 필수 소품인 쇠사슬이 누락되었고 표정에서 공포감이 부족합니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 쇠사슬 소품을 충실히 포함했으며, 공포에 질린 창백한 얼굴과 결박된 상태를 사실적이고 다양하게 잘 묘사했습니다." }, { "label": "B", "score": 4, "verdict_ko": "제시된 쇠사슬 소품이 완전히 누락되었으며, 인물들의 복장이 동일하고 표정에 공포감이 부족하여 샷 텍스트의 감정 묘사를 살리지 못했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 쇠사슬 소품을 충실히 포함했으며, 공포에 질린 창백한 얼굴과 결박된 상태를 사실적이고 다양하게 잘 묘사했습니다." }, { "label": "A", "score": 4, "verdict_ko": "제시된 쇠사슬 소품이 완전히 누락되었으며, 인물들의 복장이 동일하고 표정에 공포감이 부족하여 샷 텍스트의 감정 묘사를 살리지 못했습니다." } ] }, "combined": { "totals": { "A": 7, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 7, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 어창 배경과 쇠사슬 소품을 정확히 포함하였으며, 공포에 질려 위를 올려다보는 창백한 얼굴들을 성공적으로 연출했습니다." }, { "label": "A", "score": 3, "verdict_ko": "전경에 위치한 인물 주변에 주인을 알 수 없는 발이 나타나는 해부학적 오류가 있으며, 필수 소품인 쇠사슬이 누락되었고 표정에서 공포감이 부족합니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B05.png" }, { "label": "CHARACTER REFERENCE — 정인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 굵은 쇠사슬: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S102sh16__bgfirst_bg.png", "bg_asset_id": "c42a39ef-8a1f-424d-99a3-7b65743b53af", "bg_record_key": "S102sh16::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S102sh20::bgfirst_bg": { "input_fingerprint": "f7004957ad102862", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 순백의 무녀복을 입은 채 바위 앞으로 걸어 나오며 한 발을 내디딘 강영란(흰 머리 여자)의 압도적인 전면.\n\nLOCATION (lock): The firelit forest ritual clearing directly before the massive rock altar at evening.\n\nTIME OF DAY (lock): day transitioning to evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The frontal dolly-in ends at chest height in a dominant wide portrait, with 강영란(흰 머리 여자) centered before the rock as her forward foot completes one deliberate step into the ritual space. Her full figure and white shaman robe remain visible, while the rock rises behind her and the ritual activity is held to the peripheral background.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강영란 in the middle-center of the frame, midground, moves toward ritual space before the rock; ritual rock in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 바위 (positioned behind the ritual leader); used as Serves as the central backdrop that fixes her ceremonial entrance on the frontal axis; 제단 주변 모닥불 (burning strongly); used as Provides peripheral ritual context without competing with the advancing figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 순백의 무녀복을 입은 채 바위 앞으로 걸어 나오며 한 발을 내디딘 강영란(흰 머리 여자)의 압도적인 전면.\n\nLOCATION (lock): The firelit forest ritual clearing directly before the massive rock altar at evening.\n\nTIME OF DAY (lock): day transitioning to evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The frontal dolly-in ends at chest height in a dominant wide portrait, with 강영란(흰 머리 여자) centered before the rock as her forward foot completes one deliberate step into the ritual space. Her full figure and white shaman robe remain visible, while the rock rises behind her and the ritual activity is held to the peripheral background.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 강영란 in the middle-center of the frame, midground, moves toward ritual space before the rock; ritual rock in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 바위 (positioned behind the ritual leader); used as Serves as the central backdrop that fixes her ceremonial entrance on the frontal axis; 제단 주변 모닥불 (burning strongly); used as Provides peripheral ritual context without competing with the advancing figure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S102sh20__bgfirst_bg.png", "asset_id": "9b9d0489-b496-45a6-a72d-651d79934725", "input_asset_ids": [ "636a2bef-4e06-4740-b228-02576f0c5d7b", "cbaf7811-149d-49ab-85de-5ec1c1ad6fa4" ] }, "S102sh20": { "input_fingerprint": "8e3d7b3d61819d3a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning to evening.\n\nSHOT TEXT (authoritative, Korean): 순백의 무녀복을 입은 채 바위 앞으로 걸어 나오며 한 발을 내디딘 강영란(흰 머리 여자)의 압도적인 전면.\n\nLOCATION (lock): The firelit forest ritual clearing directly before the massive rock altar at evening. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran has changed from the red robe into a clean white shaman's robe for the main ritual. Jihu's bone necklace remains at the ceremonial site.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning to evening.\n\nSHOT TEXT (authoritative, Korean): 순백의 무녀복을 입은 채 바위 앞으로 걸어 나오며 한 발을 내디딘 강영란(흰 머리 여자)의 압도적인 전면.\n\nLOCATION (lock): The firelit forest ritual clearing directly before the massive rock altar at evening. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran has changed from the red robe into a clean white shaman's robe for the main ritual. Jihu's bone necklace remains at the ceremonial site.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day transitioning to evening.\n\nSHOT TEXT (authoritative, Korean): 순백의 무녀복을 입은 채 바위 앞으로 걸어 나오며 한 발을 내디딘 강영란(흰 머리 여자)의 압도적인 전면.\n\nLOCATION (lock): The firelit forest ritual clearing directly before the massive rock altar at evening. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran has changed from the red robe into a clean white shaman's robe for the main ritual. Jihu's bone necklace remains at the ceremonial site.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S102sh20__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S102sh20.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B04.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 인물의 외모와 복장, '한 발을 내디딘' 앞모습의 액션을 잘 구현했으며 배경의 디테일도 충실합니다." }, { "label": "B", "score": 5, "verdict_ko": "액션과 배경은 무난하나, 인물의 목 뒤로 정체불명의 막대기가 관통하는 생성 오류가 있어 감점되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 인물의 외모와 복장, '한 발을 내디딘' 앞모습의 액션을 잘 구현했으며 배경의 디테일도 충실합니다." }, { "label": "B", "score": 5, "verdict_ko": "액션과 배경은 무난하나, 인물의 목 뒤로 정체불명의 막대기가 관통하는 생성 오류가 있어 감점되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 압도적인 정면 구도와 '한 발을 내디딘' 역동적인 걷는 동작을 자연스럽게 구현하여 지시문을 가장 잘 충족합니다." }, { "label": "A", "score": 4, "verdict_ko": "걷는 동작이 제대로 표현되지 않은 정적인 자세이며, 배경에 시대와 장소에 맞지 않는 대나무 횃불 형태의 구조물이 임의로 추가되어 감점되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 압도적인 정면 구도와 '한 발을 내디딘' 역동적인 걷는 동작을 자연스럽게 구현하여 지시문을 가장 잘 충족합니다." }, { "label": "B", "score": 4, "verdict_ko": "걷는 동작이 제대로 표현되지 않은 정적인 자세이며, 배경에 시대와 장소에 맞지 않는 대나무 횃불 형태의 구조물이 임의로 추가되어 감점되었습니다." } ] }, "combined": { "totals": { "A": 15, "B": 9 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 15, "B": 9 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 인물의 외모와 복장, '한 발을 내디딘' 앞모습의 액션을 잘 구현했으며 배경의 디테일도 충실합니다." }, { "label": "B", "score": 5, "verdict_ko": "액션과 배경은 무난하나, 인물의 목 뒤로 정체불명의 막대기가 관통하는 생성 오류가 있어 감점되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L30B04.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S102sh20__bgfirst_bg.png", "bg_asset_id": "9b9d0489-b496-45a6-a72d-651d79934725", "bg_record_key": "S102sh20::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "groupbg::반대편_갯바위": { "input_fingerprint": "d19c8aa670238141", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "반대편_갯바위", "tags": [ "S103sh1", "S103sh7", "S88sh4", "S89sh1" ] }, "context_sig": "04cc5647e1fd4cf4" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: At the waterline below a rocky island shore at dusk.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n경찰 순찰정 갑판: 소형 경찰선의 외부 데크. 난간 바(rail bar)와 통로가 있다. (특징: 순찰정 외장(경찰 마크, 붉은/푸른 선); 철제 난간 바; 물에 젖은 갑판 바닥) / 금월도 갯바위: 파도가 거칠게 들이치는 날카롭고 거친 바위 지대. (특징: 파도가 부딪쳐 부서지는 하얀 포말; 검고 날카로운 거대 암석들; 바위 틈새에 고인 바닷물)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 바다를 빙- 돌아서 다시 금월도 반대쪽으로 왔던 것!\n- 실외. 금월도 갯바위 - 계속\n- 실외. 갯바위 - 저녁\n\nTIME OF DAY (lock): evening.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise: At the waterline below a rocky island shore at dusk.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n경찰 순찰정 갑판: 소형 경찰선의 외부 데크. 난간 바(rail bar)와 통로가 있다. (특징: 순찰정 외장(경찰 마크, 붉은/푸른 선); 철제 난간 바; 물에 젖은 갑판 바닥) / 금월도 갯바위: 파도가 거칠게 들이치는 날카롭고 거친 바위 지대. (특징: 파도가 부딪쳐 부서지는 하얀 포말; 검고 날카로운 거대 암석들; 바위 틈새에 고인 바닷물)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- 바다를 빙- 돌아서 다시 금월도 반대쪽으로 왔던 것!\n- 실외. 금월도 갯바위 - 계속\n- 실외. 갯바위 - 저녁\n\nTIME OF DAY (lock): evening.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_반대편_갯바위_bdbc4d.png", "asset_id": "5c4e36b5-5454-4642-9b23-510bfa7e03b3", "input_asset_ids": [ "2368c61a-3ebb-49e5-9926-c9e9bf26d757" ], "origin_tag": "S103sh1", "place_text": "At the waterline below a rocky island shore at dusk.", "origin_inputs": { "place_text": "At the waterline below a rocky island shore at dusk.", "time_of_day_en": "evening", "conti_asset_id": "2368c61a-3ebb-49e5-9926-c9e9bf26d757" } }, "S103sh1::bgfirst_bg": { "input_fingerprint": "9b0fd7fcf258d901", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 짙은 어둠이 내리는 갯바위, 철썩이는 바닷물을 뚫고 수면 위로 반쯤 솟아오른 사람의 젖은 뻗은 손.\n\nLOCATION (lock): At the waterline below a rocky island shore at dusk.\n\nTIME OF DAY (lock): evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just above water level, hold a wider frontal view of the rock edge as the camera begins its restrained push toward a single wet arm breaking the surface near lower center. Only 고산들's half-emerged hand and forearm are revealed against the surrounding water, with the descending darkness and the edge of the 갯바위 remaining clearly legible.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 고산들의 emerging hand in the lower-center of the frame, midground, moves toward edge of the coastal rock; edge of the coastal rock in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 바닷물 (striking the coastal rock in waves); used as Surrounds the emerging hand and supplies the motion that first reveals it; 갯바위 가장자리 (reached by the emerging hand); used as Provides the grip point and the boundary between the submerged body and safety.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 짙은 어둠이 내리는 갯바위, 철썩이는 바닷물을 뚫고 수면 위로 반쯤 솟아오른 사람의 젖은 뻗은 손.\n\nLOCATION (lock): At the waterline below a rocky island shore at dusk.\n\nTIME OF DAY (lock): evening.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just above water level, hold a wider frontal view of the rock edge as the camera begins its restrained push toward a single wet arm breaking the surface near lower center. Only 고산들's half-emerged hand and forearm are revealed against the surrounding water, with the descending darkness and the edge of the 갯바위 remaining clearly legible.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 고산들의 emerging hand in the lower-center of the frame, midground, moves toward edge of the coastal rock; edge of the coastal rock in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 바닷물 (striking the coastal rock in waves); used as Surrounds the emerging hand and supplies the motion that first reveals it; 갯바위 가장자리 (reached by the emerging hand); used as Provides the grip point and the boundary between the submerged body and safety.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S103sh1__bgfirst_bg.png", "asset_id": "8e81888a-79fc-40a0-8e61-513521ddf01d", "input_asset_ids": [ "2368c61a-3ebb-49e5-9926-c9e9bf26d757", "5c4e36b5-5454-4642-9b23-510bfa7e03b3" ] }, "S103sh1": { "input_fingerprint": "623c48238360ba92", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): evening.\n\nSHOT TEXT (authoritative, Korean): 짙은 어둠이 내리는 갯바위, 철썩이는 바닷물을 뚫고 수면 위로 반쯤 솟아오른 사람의 젖은 뻗은 손.\n\nLOCATION (lock): At the waterline below a rocky island shore at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul resurfaces alive with the deep back and chest stab wounds still bleeding, and his body and clothing are soaked from the sea.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): evening.\n\nSHOT TEXT (authoritative, Korean): 짙은 어둠이 내리는 갯바위, 철썩이는 바닷물을 뚫고 수면 위로 반쯤 솟아오른 사람의 젖은 뻗은 손.\n\nLOCATION (lock): At the waterline below a rocky island shore at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul resurfaces alive with the deep back and chest stab wounds still bleeding, and his body and clothing are soaked from the sea.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): evening.\n\nSHOT TEXT (authoritative, Korean): 짙은 어둠이 내리는 갯바위, 철썩이는 바닷물을 뚫고 수면 위로 반쯤 솟아오른 사람의 젖은 뻗은 손.\n\nLOCATION (lock): At the waterline below a rocky island shore at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul resurfaces alive with the deep back and chest stab wounds still bleeding, and his body and clothing are soaked from the sea.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S103sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S103sh1.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_반대편_갯바위_bdbc4d.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "숏 텍스트의 핵심 피사체인 솟아오른 손을 정확히 프레이밍했으며, 젖은 옷(소매)과 핏자국 등 명시된 상태를 잘 반영했습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레임을 넓혀 인물의 얼굴을 보여주었으나, 명시된 의상(젖은 옷)이 벗겨진 채 묘사되어 유지 상태 지침을 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "숏 텍스트의 핵심 피사체인 솟아오른 손을 정확히 프레이밍했으며, 젖은 옷(소매)과 핏자국 등 명시된 상태를 잘 반영했습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레임을 넓혀 인물의 얼굴을 보여주었으나, 명시된 의상(젖은 옷)이 벗겨진 채 묘사되어 유지 상태 지침을 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "뻗은 손만 묘사한 숏 텍스트의 의도와 달리 얼굴과 상반신까지 과도하게 보여주었으며, 젖은 옷 설정에 어긋나게 맨팔로 묘사되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "'수면 위로 솟아오른 뻗은 손'이라는 숏 텍스트의 제한된 프레이밍을 정확히 구현했으며, 피 묻은 젖은 옷의 상태도 잘 반영했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "뻗은 손만 묘사한 숏 텍스트의 의도와 달리 얼굴과 상반신까지 과도하게 보여주었으며, 젖은 옷 설정에 어긋나게 맨팔로 묘사되었습니다." }, { "label": "A", "score": 7, "verdict_ko": "'수면 위로 솟아오른 뻗은 손'이라는 숏 텍스트의 제한된 프레이밍을 정확히 구현했으며, 피 묻은 젖은 옷의 상태도 잘 반영했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 8 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 8 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "숏 텍스트의 핵심 피사체인 솟아오른 손을 정확히 프레이밍했으며, 젖은 옷(소매)과 핏자국 등 명시된 상태를 잘 반영했습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레임을 넓혀 인물의 얼굴을 보여주었으나, 명시된 의상(젖은 옷)이 벗겨진 채 묘사되어 유지 상태 지침을 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_반대편_갯바위_bdbc4d.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S103sh1__bgfirst_bg.png", "bg_asset_id": "8e81888a-79fc-40a0-8e61-513521ddf01d", "bg_record_key": "S103sh1::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "반대편_갯바위", "groupbg_asset_id": "5c4e36b5-5454-4642-9b23-510bfa7e03b3" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S103sh7": { "input_fingerprint": "37df5ab451d6ea79", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): evening.\n\nSHOT TEXT (authoritative, Korean): 갯바위에 누운 채 주먹이 바닥에 쾅 닿은 충격의 순간, 서럽게 오열하며 일그러진 고산들의 처절한 옆얼굴.\n\nLOCATION (lock): A flat wave-washed ledge on the island’s rocky coast as darkness falls. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same darkening rocks, breaking surf, wet surfaces, and evening light. Exclude the emerging-hand moment and show the soaked man collapsed on the rocks in grief.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul lies soaked on the rocks with the stab wounds in his back and chest and any resulting blood still visible.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 6, "C": 4 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "짧게 깎은 머리 등 인물 외형을 잘 반영했으며, 주먹을 내리치는 충격과 오열하는 옆얼굴을 가장 정확하게 묘사했습니다." }, { "label": "B", "score": 6, "verdict_ko": "오열하는 표정과 배경은 훌륭하나, 지시된 '짧게 깎은 머리'에 비해 머리가 길고 타격의 순간이 다소 약하게 표현되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "주먹을 쥔 자세와 배경은 일치하나, 요구된 '오열하며 일그러진 옆얼굴'이 대부분 가려져 감정 표현이 제대로 드러나지 않습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S103sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S103sh1" } }, "S104sh4::bgfirst_bg": { "input_fingerprint": "69444fb831c60fb4", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 흠뻑 젖은 모습으로 살기 어린 눈빛을 띤 채, 마을 길을 걷는 중간 한 발이 공중에 떠 있는 고산들의 위압적인 전면.\n\nLOCATION (lock): The village lane directly in front of the restaurant at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on the road-to-entrance axis, retreat directly ahead of 고산들 in a tightened full-body frame, placing his hostile face in the upper center and his lifted stepping foot just above the lower edge. His soaked body advances squarely into the lens path while the restaurant entrance remains behind the camera, matching his fixed line of attack toward 황봉업.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 고산들 in the middle-center of the frame, midground, moves toward restaurant entrance and 황봉업.\n- KEY BACKGROUND ELEMENTS: 식당으로 향하는 마을길 (occupied by the approaching figure); used as Forms the receding corridor behind the advancing figure and clarifies his approach toward the restaurant; 반월 식당 입구 (being used by 황봉업 and 구창모 to flee inside) — The entrance faces the approaching 고산들 from behind the camera axis; used as Defines the destination of the approach just beyond the camera's retreating position.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe Banwol Restaurant main building is the organizing mass of the site. Its front entrance opens onto the restaurant-front alley, which joins the village-road approach aligned toward the entrance. From the restaurant side, the village road also continues uphill until it reaches the forest and mountain threshold. Behind the main building, the back entrance opens directly into the restaurant rear yard. The yard contains a cultivated garden plot and the separate storage shed, with open circulation linking the back entrance, garden edge, and shed door. The shed door faces into the yard so cameras can work from the yard or directly at the shed threshold. A perimeter boundary defines the rear-yard extent while preserving its connections to the building and village-road side.\n- building: Banwol Restaurant main building\n- entrance: Restaurant front entrance\n- alley: Restaurant-front alley\n- approach: Restaurant-approach village road\n- yard: Restaurant rear yard\n- garden: Cultivated rear-yard garden plot\n- entrance: Restaurant back entrance\n- boundary: Rear-yard perimeter boundary\n- shed: Rear-yard storage shed\n- door: Rear-yard storage shed door\n- road: Uphill village-road continuation\n- boundary: Forest and mountain threshold\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 흠뻑 젖은 모습으로 살기 어린 눈빛을 띤 채, 마을 길을 걷는 중간 한 발이 공중에 떠 있는 고산들의 위압적인 전면.\n\nLOCATION (lock): The village lane directly in front of the restaurant at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on the road-to-entrance axis, retreat directly ahead of 고산들 in a tightened full-body frame, placing his hostile face in the upper center and his lifted stepping foot just above the lower edge. His soaked body advances squarely into the lens path while the restaurant entrance remains behind the camera, matching his fixed line of attack toward 황봉업.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 고산들 in the middle-center of the frame, midground, moves toward restaurant entrance and 황봉업.\n- KEY BACKGROUND ELEMENTS: 식당으로 향하는 마을길 (occupied by the approaching figure); used as Forms the receding corridor behind the advancing figure and clarifies his approach toward the restaurant; 반월 식당 입구 (being used by 황봉업 and 구창모 to flee inside) — The entrance faces the approaching 고산들 from behind the camera axis; used as Defines the destination of the approach just beyond the camera's retreating position.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe Banwol Restaurant main building is the organizing mass of the site. Its front entrance opens onto the restaurant-front alley, which joins the village-road approach aligned toward the entrance. From the restaurant side, the village road also continues uphill until it reaches the forest and mountain threshold. Behind the main building, the back entrance opens directly into the restaurant rear yard. The yard contains a cultivated garden plot and the separate storage shed, with open circulation linking the back entrance, garden edge, and shed door. The shed door faces into the yard so cameras can work from the yard or directly at the shed threshold. A perimeter boundary defines the rear-yard extent while preserving its connections to the building and village-road side.\n- building: Banwol Restaurant main building\n- entrance: Restaurant front entrance\n- alley: Restaurant-front alley\n- approach: Restaurant-approach village road\n- yard: Restaurant rear yard\n- garden: Cultivated rear-yard garden plot\n- entrance: Restaurant back entrance\n- boundary: Rear-yard perimeter boundary\n- shed: Rear-yard storage shed\n- door: Rear-yard storage shed door\n- road: Uphill village-road continuation\n- boundary: Forest and mountain threshold\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S104sh4__bgfirst_bg.png", "asset_id": "253adbf0-f5a2-4224-96b7-8957c1b81ef0", "input_asset_ids": [ "25cc7574-4dd0-46e4-b58f-777d103f3d1b" ] }, "S104sh4": { "input_fingerprint": "ebbb511f2af17f9b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 흠뻑 젖은 모습으로 살기 어린 눈빛을 띤 채, 마을 길을 걷는 중간 한 발이 공중에 떠 있는 고산들의 위압적인 전면.\n\nLOCATION (lock): The village lane directly in front of the restaurant at night. The shot takes place here — the attached STORYBOARD SKETCH fixes the staging, camera and figure placement of this exact place. No location photograph is attached — build the location itself strictly from the location text above and the shot text, inventing nothing beyond them.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul remains soaked and visibly wounded from Youngran's stabbing as he advances toward Hwang and Changmo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 흠뻑 젖은 모습으로 살기 어린 눈빛을 띤 채, 마을 길을 걷는 중간 한 발이 공중에 떠 있는 고산들의 위압적인 전면.\n\nLOCATION (lock): The village lane directly in front of the restaurant at night. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul remains soaked and visibly wounded from Youngran's stabbing as he advances toward Hwang and Changmo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 흠뻑 젖은 모습으로 살기 어린 눈빛을 띤 채, 마을 길을 걷는 중간 한 발이 공중에 떠 있는 고산들의 위압적인 전면.\n\nLOCATION (lock): The village lane directly in front of the restaurant at night. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul remains soaked and visibly wounded from Youngran's stabbing as he advances toward Hwang and Changmo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 흠뻑 젖은 모습으로 살기 어린 눈빛을 띤 채, 마을 길을 걷는 중간 한 발이 공중에 떠 있는 고산들의 위압적인 전면.\n\nLOCATION (lock): The village lane directly in front of the restaurant at night. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul remains soaked and visibly wounded from Youngran's stabbing as he advances toward Hwang and Changmo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S104sh4__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S104sh4__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S104sh4__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "totals": { "A": 8, "B": 6, "C": 5 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "스토리보드와 정확히 일치하는 걷는 포즈(한 발이 뜬 상태)를 구현했으며, 흠뻑 젖은 의상, 상처, 살기 어린 표정까지 프롬프트의 지시를 가장 완벽하게 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "캐릭터와 배경의 일치도는 높으나, 걷는 모습이 다소 뛰는 듯한 자세로 연출되어 스토리보드의 정적이고 위압적인 포즈와 차이가 있습니다." }, { "label": "C", "score": 5, "verdict_ko": "전체적인 구도는 맞으나 자세가 다소 경직되어 있으며, 들려 있는 발의 각도와 손가락의 해부학적 묘사가 부자연스럽습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S104sh4__bgfirst_bg.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S104sh4__bgfirst_bg.png", "bg_asset_id": "253adbf0-f5a2-4224-96b7-8957c1b81ef0", "bg_record_key": "S104sh4::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+엔티티", "share_plan": { "ref_plan": "background" } }, "S104sh8": { "input_fingerprint": "07e66850738ff7e5", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 맹렬한 기세로 허공을 향해 뛰어올라 두 발이 바닥에서 완전히 떨어진 상태인 고산들의 역동적인 실루엣.\n\nLOCATION (lock): The open area immediately outside the restaurant entrance. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same restaurant exterior, narrow lane, wet night surfaces, and darkness. Exclude the man's walking stride and show him launching fully into the air toward the fleeing men.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul remains sea-soaked and wounded as he leaps toward the restaurant.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "샷 텍스트에 묘사된 역동적인 도약 자세를 훌륭하게 구현했으며, 추가 인물 금지 지침을 완벽히 준수했습니다." }, { "label": "B", "score": 5, "verdict_ko": "두 발이 공중에 떠 있으나, 요구된 '맹렬한 기세'나 '역동적인 실루엣'이 느껴지지 않고 다소 경직된 자세입니다." }, { "label": "C", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 인물들(도망치는 남자들)을 화면에 추가하여 등장인물 제한 지침을 심각하게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S104sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S104sh4" } }, "S105sh1::bgfirst_bg": { "input_fingerprint": "d57526e3f1bed5ae", "prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 숲속 도당, 활활 치솟는 커다란 모닥불 주변을 둥글게 둘러싼 수십 명의 한국인 검은 옷 신도들의 압도적인 전경.\n\nLOCATION (lock): The full forest ritual ground, with worshippers encircling tall fires and the massive rock altar.\n\nTIME OF DAY (lock): night, full moon, lunar eclipse beginning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Crane high above the outer edge of the gathering and angle steeply downward, revealing the complete circular formation of Korean black-clad devotees around the large bonfire. The fire occupies less than the central third while the human ring and surrounding ritual ground spread across the wider frame, emphasizing collective scale and isolation.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 커다란 모닥불 (burning high); used as Acts as the central focal point and organizes the devotees into a complete circle; 숲속 만월제 도당 (occupied by the circular gathering); used as Defines the ceremonial layout around the central fire from the overhead view.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is an open ritual clearing enclosed on every side by forest. A towering giant rock occupies the rear of the principal ritual forecourt and faces into the open ground. The central altar stands immediately in front of the rock on the clearing side, creating the main spatial axis: outer clearing, altar, then giant rock. Open ground extends laterally around this forecourt and outward to the surrounding forest edge, allowing movement around the rock and assembly across the wider clearing. A traversable forest-side approach enters the outer clearing through the enclosing woodland, although the source does not establish its compass direction. No permanent fire structures are specified; bonfires, smoke, blood traces, restraints, offerings, equipment, and participant formations remain moment-specific additions rather than site markers.\n- clearing: forest-enclosed ritual clearing\n- rock: towering giant ritual rock\n- altar: central altar before the giant rock\n- boundary: surrounding forest edge\n- approach: traversable forest-side approach\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "effective_prompt": "Turn the attached storyboard sketch into a photorealistic film still of\nits LOCATION ONLY, keeping the figures exactly as they are.\n\nKEEP EXACTLY: the camera framing, the horizon, and where every element\nsits in the frame. Each mannequin figure stays a plain grey featureless\nmannequin standing in the very same spot, at the same size, in the same\npose, turned the same way — do not turn them into people, do not move,\nrotate, mirror or re-pose them, do not add or remove figures.\n\nBUILD PHOTOREALISTICALLY: everything that is not a figure — ground,\nsurfacing, structures, vegetation, sky, water, distance. The bare lines\nof the sketch are a layout guide; replace them with the real materials,\ndepth and lighting of the place described below.\n\nSHOT TEXT this background must serve (Korean): 숲속 도당, 활활 치솟는 커다란 모닥불 주변을 둥글게 둘러싼 수십 명의 한국인 검은 옷 신도들의 압도적인 전경.\n\nLOCATION (lock): The full forest ritual ground, with worshippers encircling tall fires and the massive rock altar.\n\nTIME OF DAY (lock): night, full moon, lunar eclipse beginning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Crane high above the outer edge of the gathering and angle steeply downward, revealing the complete circular formation of Korean black-clad devotees around the large bonfire. The fire occupies less than the central third while the human ring and surrounding ritual ground spread across the wider frame, emphasizing collective scale and isolation.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 커다란 모닥불 (burning high); used as Acts as the central focal point and organizes the devotees into a complete circle; 숲속 만월제 도당 (occupied by the circular gathering); used as Defines the ceremonial layout around the central fire from the overhead view.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nTHINGS AT THIS PLACE:\nThe site is an open ritual clearing enclosed on every side by forest. A towering giant rock occupies the rear of the principal ritual forecourt and faces into the open ground. The central altar stands immediately in front of the rock on the clearing side, creating the main spatial axis: outer clearing, altar, then giant rock. Open ground extends laterally around this forecourt and outward to the surrounding forest edge, allowing movement around the rock and assembly across the wider clearing. A traversable forest-side approach enters the outer clearing through the enclosing woodland, although the source does not establish its compass direction. No permanent fire structures are specified; bonfires, smoke, blood traces, restraints, offerings, equipment, and participant formations remain moment-specific additions rather than site markers.\n- clearing: forest-enclosed ritual clearing\n- rock: towering giant ritual rock\n- altar: central altar before the giant rock\n- boundary: surrounding forest edge\n- approach: traversable forest-side approach\n\nWORLD FACTS (creator-confirmed — always true):\n- Region (real-world reference): 대한민국의 대도시권, 남서해안권 및 외딴 한국 섬 지역; 역사 구간은 조선과 식민지기 한반도\n- Era: 주된 배경은 2020년대 중반의 현대이며, 17세기 조선과 일제강점기 전후를 다루는 역사적 재현 구간이 병존함\n- 현대 장면에는 동시대 대한민국의 생활 환경을 적용하고, 17세기 조선 재현에는 초가·한복·갓·전근대 생활도구를, 식민지기 재현에는 당시 조선인 복식과 일본군 제식 장비를 사용한다. 서로 다른 시대의 복식·무기·교통수단·전자기기를 혼용하지 않는다.\n- 현대 인물에게 스마트폰, 지도 앱, CCTV, 바코드 출근 시스템, 현대 경찰 장비와 차량을 사용할 수 있다. 외딴 지역에서는 신호 두절과 구형 SUV·소형 어선·순찰정 같은 제한적 기반시설을 반영하되, 역사 재현 구간에는 현대 전자기기와 현대식 차량을 배제한다.\n\nTHINGS THAT LIVE AT THIS PLACE (what they are, not where they go): the\nlist above names what permanently belongs to this site. Where each one\nsits in this frame is decided by the attached sketch alone — do not add\nor reposition anything on the strength of the list, build only what the\nsketch already has lines for, and do not invent facilities that are not\nlisted.\n\nNo text, no captions, no diagram symbols, no markers anywhere in the\nimage.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S105sh1__bgfirst_bg.png", "asset_id": "72ac7ae7-eb3a-4d30-ac0b-c4a6d03d0490", "input_asset_ids": [ "0c31ccc2-93aa-4315-83c2-a09908ef7409" ] }, "S105sh1": { "input_fingerprint": "8f282af62fb582ff", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, full moon, lunar eclipse beginning.\n\nSHOT TEXT (authoritative, Korean): 숲속 도당, 활활 치솟는 커다란 모닥불 주변을 둥글게 둘러싼 수십 명의 한국인 검은 옷 신도들의 압도적인 전경.\n\nLOCATION (lock): The full forest ritual ground, with worshippers encircling tall fires and the massive rock altar. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same giant rock, circular bonfires, smoky clearing, and surrounding dark forest. Exclude the earlier sparse attendants and fill the circle with black-clad worshippers.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The villagers remain dressed in black ceremonial clothing around the fires, while Youngran wears the white shaman's robe. Jihu's necklace remains at the altar.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, full moon, lunar eclipse beginning.\n\nSHOT TEXT (authoritative, Korean): 숲속 도당, 활활 치솟는 커다란 모닥불 주변을 둥글게 둘러싼 수십 명의 한국인 검은 옷 신도들의 압도적인 전경.\n\nLOCATION (lock): The full forest ritual ground, with worshippers encircling tall fires and the massive rock altar. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The villagers remain dressed in black ceremonial clothing around the fires, while Youngran wears the white shaman's robe. Jihu's necklace remains at the altar.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, full moon, lunar eclipse beginning.\n\nSHOT TEXT (authoritative, Korean): 숲속 도당, 활활 치솟는 커다란 모닥불 주변을 둥글게 둘러싼 수십 명의 한국인 검은 옷 신도들의 압도적인 전경.\n\nLOCATION (lock): The full forest ritual ground, with worshippers encircling tall fires and the massive rock altar. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The villagers remain dressed in black ceremonial clothing around the fires, while Youngran wears the white shaman's robe. Jihu's necklace remains at the altar.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "C": "Replace every grey or outlined mannequin figure in the FIRST attached\nimage with the real people the CHARACTER REFERENCE images show, and\noutput a photorealistic film still. Each mannequin becomes the character\nwhose POSE CANON and the shot text match that figure's pose and\nposition; never repeat one person across two mannequins. If no CHARACTER\nREFERENCE image is attached for a figure, still replace it with a\nplausible real person consistent with the text. Every mannequin becomes\na person — erasing a figure, or leaving one out, is not a replacement,\nand the number of figures never changes.\n\nKEEP EXACTLY: the background of the first image, the camera framing,\nand each mannequin's position, size, pose and the direction it is\nturned. A person must be exactly where their mannequin was, at the same\nscale, in the same pose, facing the same way — never mirrored, never\nre-staged, never straightened up, stood upright or otherwise re-posed.\nNo mannequin, grey figure or sketch line may remain anywhere in the\noutput.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, full moon, lunar eclipse beginning.\n\nSHOT TEXT (authoritative, Korean): 숲속 도당, 활활 치솟는 커다란 모닥불 주변을 둥글게 둘러싼 수십 명의 한국인 검은 옷 신도들의 압도적인 전경.\n\nLOCATION (lock): The full forest ritual ground, with worshippers encircling tall fires and the massive rock altar. The shot takes place here — the FIRST attached image (SHOT BACKGROUND) is this exact place, already built: its ground, structures, horizon, materials and lighting are the finished truth of this location and must not be redesigned or replaced. No location photograph is attached — read the place from that image alone, and add no scenery, structure, vehicle or fixture that it does not already show. This lock governs the place only; the figures in the shot follow the staging and pose instructions.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The villagers remain dressed in black ceremonial clothing around the fires, while Youngran wears the white shaman's robe. Jihu's necklace remains at the altar.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S105sh1__bgfirst_bg.png" } ], "B": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S105sh1__bgfirst_bg.png" } ], "C": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S105sh1__bgfirst_bg.png" } ] }, "totals": { "B": 7, "C": 5, "A": 3 }, "selected": "B", "ranking": [ "B", "C", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "레퍼런스의 제단 앞 인물 위치와 두 팔을 벌린 자세를 완벽히 유지했으며, 검은 옷 신도들과 흰 옷 무당의 구성을 가장 정확히 구현했습니다." }, { "label": "C", "score": 5, "verdict_ko": "월식의 표현은 반영되었으나, 제단 앞을 지켜야 할 흰 옷 무당의 위치가 프레임 하단으로 이동하여 레퍼런스의 연속성 유지에 실패했습니다." }, { "label": "A", "score": 3, "verdict_ko": "제단 앞 인물이 누락되었고 흰 옷을 입은 인물이 두 명으로 중복 생성되어 인물 구성 및 연속성 지시를 크게 위반했습니다." } ], "refs": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S105sh1__bgfirst_bg.png" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S105sh1__bgfirst_bg.png", "bg_asset_id": "72ac7ae7-eb3a-4d30-ac0b-c4a6d03d0490", "bg_record_key": "S105sh1::bgfirst_bg", "chain_winner": true, "authority": "lane_conti_only" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S102sh20" } }, "S105sh6": { "input_fingerprint": "026c9ec2c884fb45", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, full moon, lunar eclipse beginning.\n\nSHOT TEXT (authoritative, Korean): 피로 물든 무녀복을 입은 채 광기 어린 미소를 지으며 제단 앞에 우뚝 선 강영란(흰 머리 여자)의 기괴한 전면.\n\nLOCATION (lock): The central altar space before the great rock, illuminated by blazing ritual fires and the full moon. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same towering fires, ring of worshippers, giant rock, and eclipse-night atmosphere. Exclude the wide crowd as the main subject and show the elder before the altar in blood-soaked ritual clothes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Youngran's formerly white shaman's robe is now heavily stained with fresh goat blood, which must remain through the ritual. The rock and altar are also freshly splattered, and Jihu's necklace remains there.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "B": 5, "A": 4 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "목걸이를 유일하게 묘사하고 참조된 제단 형태를 유지했으나, 전경에 제단(돌)이 중복 생성되는 공간적 오류가 있습니다." }, { "label": "B", "score": 5, "verdict_ko": "표정은 우수하나, 지시문에 없는 죽은 염소를 임의로 생성(심각한 위반)하고 제단의 형태를 변형했습니다." }, { "label": "A", "score": 4, "verdict_ko": "제단을 흰 천으로 덮어 장소 일관성을 위반했고, 바닥에 임의의 사물들이 추가되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S105sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S105sh1" } }, "S106sh2": { "input_fingerprint": "e75ea725c487bfdf", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 흰 사슴뿔 창을 한 손에 꽉 쥔 채 어두운 식당 밖으로 막 걸음을 내딛는 피범벅 된 고산들의 전신.\n\nLOCATION (lock): The restaurant entrance as a blood-covered figure steps from the interior onto the dark village lane. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark restaurant entrance, village lane, wet surfaces, and moonlight. Exclude the earlier unarmed approach and show the blood-covered man exiting with the white-antler spear; remove the living boat operator and leave his body inside.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): His blood-soaked corpse is sprawled motionless across the restaurant hall floor behind 고산들, with his torso supported by the floor and his head and limbs collapsed loosely around it.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul remains soaked, wounded, and covered in blood while carrying the old white antler-tipped spear taken from the shed altar. Hwang's blood-soaked corpse is sprawled motionless across the restaurant hall floor behind Sandeul, with his torso supported by the floor and his head and limbs collapsed loosely around it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "B": 7, "A": 5, "C": 3 }, "selected": "B", "ranking": [ "B", "A", "C" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "시신이 식당 내부에 있지 않고 밖에 배치된 점은 아쉽지만, 창의 올바른 방향과 골목의 원래 지형을 가장 잘 유지했습니다." }, { "label": "A", "score": 5, "verdict_ko": "시신이 외부에 위치하며, 사슴뿔 창을 거꾸로 쥐고 있어 소품 사용과 묘사 지침을 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "레퍼런스에 없는 계단을 골목에 임의로 생성하여 장소 고정(Location lock) 지침을 크게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S104sh8_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 사슴뿔 창: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S104sh8" }, "lane_policy": "ab_select_bypass:prev" }, "S107sh1::bgfirst_bg": { "input_fingerprint": "25e3bb8c3534e03c", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 칠흑 같은 숲속에서 튀어나온 나뭇가지를 한 팔로 거칠게 쳐낸 채, 상체를 앞으로 강하게 숙이며 질주하는 도약 자세의 땀 범벅된 구창모의 상체.\n\nLOCATION (lock): A trackless, pitch-dark section of dense maritime-pine forest.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track backward at upper-torso height, close ahead and slightly to 구창모’s front-left, keeping his forward-pitched upper body across the center and right half of frame. His sweeping arm crosses the near foreground as it knocks aside a branch, briefly obscuring part of his sweat-covered face while the running momentum remains legible behind it.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 나뭇가지 (Being knocked aside) — The branch projects diagonally toward 구창모’s path and is being struck sideways; used as A branch enters the near foreground and gives the arm sweep a visible obstacle; 수풀 (Crowding the running path); used as Compressed behind and beside the runner to preserve the sense of an unreadable route; 숲길 (Almost indistinguishable in the darkness) — The route recedes behind 구창모 without a clearly readable direction; used as Forms the receding spatial axis behind 구창모 while remaining difficult to distinguish.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 칠흑 같은 숲속에서 튀어나온 나뭇가지를 한 팔로 거칠게 쳐낸 채, 상체를 앞으로 강하게 숙이며 질주하는 도약 자세의 땀 범벅된 구창모의 상체.\n\nLOCATION (lock): A trackless, pitch-dark section of dense maritime-pine forest.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track backward at upper-torso height, close ahead and slightly to 구창모’s front-left, keeping his forward-pitched upper body across the center and right half of frame. His sweeping arm crosses the near foreground as it knocks aside a branch, briefly obscuring part of his sweat-covered face while the running momentum remains legible behind it.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 나뭇가지 (Being knocked aside) — The branch projects diagonally toward 구창모’s path and is being struck sideways; used as A branch enters the near foreground and gives the arm sweep a visible obstacle; 수풀 (Crowding the running path); used as Compressed behind and beside the runner to preserve the sense of an unreadable route; 숲길 (Almost indistinguishable in the darkness) — The route recedes behind 구창모 without a clearly readable direction; used as Forms the receding spatial axis behind 구창모 while remaining difficult to distinguish.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S107sh1__bgfirst_bg.png", "asset_id": "b0d0c38b-9b59-41c1-922c-96fa25890363", "input_asset_ids": [ "2b4c1e9a-54d4-41c5-adbd-af1245a2a5ba", "dded1895-8314-4902-9d6c-855ef980a51c" ] }, "S107sh1": { "input_fingerprint": "ab82691364dddda5", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 칠흑 같은 숲속에서 튀어나온 나뭇가지를 한 팔로 거칠게 쳐낸 채, 상체를 앞으로 강하게 숙이며 질주하는 도약 자세의 땀 범벅된 구창모의 상체.\n\nLOCATION (lock): A trackless, pitch-dark section of dense maritime-pine forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hwang remains dead and bloodied on the restaurant floor, while Changmo flees without him.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 칠흑 같은 숲속에서 튀어나온 나뭇가지를 한 팔로 거칠게 쳐낸 채, 상체를 앞으로 강하게 숙이며 질주하는 도약 자세의 땀 범벅된 구창모의 상체.\n\nLOCATION (lock): A trackless, pitch-dark section of dense maritime-pine forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hwang remains dead and bloodied on the restaurant floor, while Changmo flees without him.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 칠흑 같은 숲속에서 튀어나온 나뭇가지를 한 팔로 거칠게 쳐낸 채, 상체를 앞으로 강하게 숙이며 질주하는 도약 자세의 땀 범벅된 구창모의 상체.\n\nLOCATION (lock): A trackless, pitch-dark section of dense maritime-pine forest. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hwang remains dead and bloodied on the restaurant floor, while Changmo flees without him.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S107sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S107sh1.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "이미지에 화살표 마커가 유출되었고 인물 등 뒤에 불필요한 팔이 생성되어 치명적인 규정 위반입니다." }, { "label": "B", "score": 7, "verdict_ko": "어두운 숲속에서 나뭇가지를 쳐내며 질주하는 구창모의 상체와 땀에 젖은 모습을 지시문과 레퍼런스에 맞게 충실히 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "이미지에 화살표 마커가 유출되었고 인물 등 뒤에 불필요한 팔이 생성되어 치명적인 규정 위반입니다." }, { "label": "B", "score": 7, "verdict_ko": "어두운 숲속에서 나뭇가지를 쳐내며 질주하는 구창모의 상체와 땀에 젖은 모습을 지시문과 레퍼런스에 맞게 충실히 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물의 외형, 땀에 젖어 질주하는 동작, 배경이 지시문과 정확히 일치하며 위반 사항이 없습니다." }, { "label": "B", "score": 3, "verdict_ko": "화면에 화살표 마커가 유출되었고 인물 뒤로 기형적인 팔이 생겨나 치명적인 규정 위반입니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "인물의 외형, 땀에 젖어 질주하는 동작, 배경이 지시문과 정확히 일치하며 위반 사항이 없습니다." }, { "label": "A", "score": 3, "verdict_ko": "화면에 화살표 마커가 유출되었고 인물 뒤로 기형적인 팔이 생겨나 치명적인 규정 위반입니다." } ] }, "combined": { "totals": { "A": 6, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "이미지에 화살표 마커가 유출되었고 인물 등 뒤에 불필요한 팔이 생성되어 치명적인 규정 위반입니다." }, { "label": "B", "score": 7, "verdict_ko": "어두운 숲속에서 나뭇가지를 쳐내며 질주하는 구창모의 상체와 땀에 젖은 모습을 지시문과 레퍼런스에 맞게 충실히 구현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/groupbg_해송_숲길_89ff8c.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S107sh1__bgfirst_bg.png", "bg_asset_id": "b0d0c38b-9b59-41c1-922c-96fa25890363", "bg_record_key": "S107sh1::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "해송_숲길", "groupbg_asset_id": "dded1895-8314-4902-9d6c-855ef980a51c" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S108sh4::bgfirst_bg": { "input_fingerprint": "a159411f21bf1808", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 창고 안쪽 바닥에 웅크린 채 양손으로 입을 꽉 틀어막고 눈물을 글썽이는 구창모의 굽은 상체.\n\nLOCATION (lock): The cramped inner storeroom of the dark slaughterhouse, behind a closed side door.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Place the static camera near floor height in front of 구창모, tilted gently upward so his folded torso occupies the middle of frame while his hands seal the lower half of his face. Keep the storage-room edges visible around him, with the side-door area held near one frame edge rather than allowing the close observation to become an isolated portrait.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 창고 안쪽 (Dark); used as Provides the enclosing background around 구창모’s crouched body; 창고 쪽문 (구창모 is hiding beside it; exact open or closed position is unspecified) — Its inner side and handle area face the camera obliquely; used as Held at the edge of frame as the fragile boundary between hiding place and danger outside; 창고 바닥 (Visible beneath the crouching figure); used as Establishes the camera’s low placement and grounds 구창모’s compressed posture.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 창고 안쪽 바닥에 웅크린 채 양손으로 입을 꽉 틀어막고 눈물을 글썽이는 구창모의 굽은 상체.\n\nLOCATION (lock): The cramped inner storeroom of the dark slaughterhouse, behind a closed side door.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Place the static camera near floor height in front of 구창모, tilted gently upward so his folded torso occupies the middle of frame while his hands seal the lower half of his face. Keep the storage-room edges visible around him, with the side-door area held near one frame edge rather than allowing the close observation to become an isolated portrait.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 창고 안쪽 (Dark); used as Provides the enclosing background around 구창모’s crouched body; 창고 쪽문 (구창모 is hiding beside it; exact open or closed position is unspecified) — Its inner side and handle area face the camera obliquely; used as Held at the edge of frame as the fragile boundary between hiding place and danger outside; 창고 바닥 (Visible beneath the crouching figure); used as Establishes the camera’s low placement and grounds 구창모’s compressed posture.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S108sh4__bgfirst_bg.png", "asset_id": "84ad65b2-ade8-4206-abd7-c16567f84046", "input_asset_ids": [ "f1eb88b5-f4ee-47e6-ae3b-6530d11a1e46", "b827b397-677b-4d45-a82e-2d3248f07276" ] }, "S108sh4": { "input_fingerprint": "114d0abc614ae8c3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 창고 안쪽 바닥에 웅크린 채 양손으로 입을 꽉 틀어막고 눈물을 글썽이는 구창모의 굽은 상체.\n\nLOCATION (lock): The cramped inner storeroom of the dark slaughterhouse, behind a closed side door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Changmo remains hidden inside the slaughterhouse side room, gripping the door and trying to suppress his breathing.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 창고 안쪽 바닥에 웅크린 채 양손으로 입을 꽉 틀어막고 눈물을 글썽이는 구창모의 굽은 상체.\n\nLOCATION (lock): The cramped inner storeroom of the dark slaughterhouse, behind a closed side door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Changmo remains hidden inside the slaughterhouse side room, gripping the door and trying to suppress his breathing.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 창고 안쪽 바닥에 웅크린 채 양손으로 입을 꽉 틀어막고 눈물을 글썽이는 구창모의 굽은 상체.\n\nLOCATION (lock): The cramped inner storeroom of the dark slaughterhouse, behind a closed side door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Changmo remains hidden inside the slaughterhouse side room, gripping the door and trying to suppress his breathing.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 구창모 (40대 중후반 한국인 남성, 가무잡잡한 얼굴, 호리한 인상, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S108sh4__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/conti/conti_S108sh4.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B04.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "샷 텍스트에 명시된 '눈물을 글썽이는' 표정과 양손으로 입을 꽉 막고 웅크린 상체의 묘사를 매우 정확하고 사실적으로 구현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "양손으로 입을 막고 웅크린 자세와 배경의 구현은 우수하나, 샷 텍스트에서 요구한 '눈물을 글썽이는' 디테일이 시각적으로 드러나지 않습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "샷 텍스트에 명시된 '눈물을 글썽이는' 표정과 양손으로 입을 꽉 막고 웅크린 상체의 묘사를 매우 정확하고 사실적으로 구현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "양손으로 입을 막고 웅크린 자세와 배경의 구현은 우수하나, 샷 텍스트에서 요구한 '눈물을 글썽이는' 디테일이 시각적으로 드러나지 않습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "양손으로 입을 꽉 틀어막은 동작과 눈물을 흘리는 표정을 텍스트의 묘사대로 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "양손으로 입을 꽉 틀어막지 않고 턱 주변에 모으고만 있으며, 눈물을 글썽이는 표정 묘사도 부족합니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "양손으로 입을 꽉 틀어막은 동작과 눈물을 흘리는 표정을 텍스트의 묘사대로 정확하게 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "양손으로 입을 꽉 틀어막지 않고 턱 주변에 모으고만 있으며, 눈물을 글썽이는 표정 묘사도 부족합니다." } ] }, "combined": { "totals": { "A": 13, "B": 18 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 13, "B": 18 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "샷 텍스트에 명시된 '눈물을 글썽이는' 표정과 양손으로 입을 꽉 막고 웅크린 상체의 묘사를 매우 정확하고 사실적으로 구현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "양손으로 입을 막고 웅크린 자세와 배경의 구현은 우수하나, 샷 텍스트에서 요구한 '눈물을 글썽이는' 디테일이 시각적으로 드러나지 않습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/episodes/18f15068-4349-4818-82a9-3b9b173110e0/images/background_chain/L37B04.png" }, { "label": "CHARACTER REFERENCE — 구창모: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S108sh4__bgfirst_bg.png", "bg_asset_id": "84ad65b2-ade8-4206-abd7-c16567f84046", "bg_record_key": "S108sh4::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S108sh8": { "input_fingerprint": "77ccc55a77bf6717", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 창고 천장에 기괴한 자세로 거꾸로 매달려 아래를 노려보는 피칠갑을 한 고산들의 서늘한 정면.\n\nLOCATION (lock): The shadowed ceiling space above the slaughterhouse’s inner storeroom. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cramped storage recess, dark ceiling, blood drips, and minimal night light. Exclude the officer as the sole subject and reveal the blood-covered man hanging directly above him.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Sandeul remains soaked, stab-wounded, and blood-covered as he hangs from the ceiling above Changmo; drops of his blood fall onto Changmo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 고산들 (40대 중반 한국인 남성, 차가운 얼굴, 날카로운 눈매, 짧게 깎은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 5, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "피칠갑 상태, 거꾸로 매달린 기괴한 자세, 창고 천장 배경 및 캐릭터의 기본 의상(회색 스웨터)을 가장 정확하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "유혈 상태와 구도는 적절하나, 지정된 캐릭터 레퍼런스와 일치하지 않는 밝은 셔츠 의상을 입고 있어 감점되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "의상과 배경은 일치하지만, 텍스트에서 강하게 요구한 '피칠갑을 한' 상태가 완전히 누락되어 치명적인 감점 요소가 되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S108sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 고산들: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S108sh4" } }, "S109sh1": { "input_fingerprint": "36c00842a9cff2fd", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 낡은 염소 축사 안, 어둠 속에서 번뜩이는 눈으로 허공을 향해 일제히 기괴하게 울부짖는 수십 마리의 검은 염소 떼 전경.\n\nLOCATION (lock): Inside the open-sided goat enclosure in the forest at night. The shot takes place here — the attached LOCATION STRUCTURE PHOTOGRAPH is the single authority for this exact place — its fixed structure and permanent site details are LOCKED to it. No separate location photograph exists for this place. Build everything else strictly from the location text above and the shot text; the layout sketch (when attached) governs framing and placement only, and the shot text governs time of day, lighting and action.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The goats remain confined inside the enclosure while Changmo is being attacked in the nearby slaughterhouse.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 7, "A": 4, "B": 4 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "번뜩이는 눈으로 허공을 향해 일제히 울부짖는 염소 떼의 핵심 행동과 야간의 축사 내부 공간을 가장 정확하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "야간 축사의 전경은 잘 담았으나, 염소들이 일제히 위를 향해 울부짖는다는 지시문 상의 핵심 행동이 누락되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "지정된 장소와 시간대는 준수했지만, 염소들이 기괴하게 울부짖는 역동적인 행동 묘사가 부족하여 지시문 충실도가 떨어집니다." } ], "refs": [ { "label": "LOCATION STRUCTURE PHOTOGRAPH — the confirmed photograph of this exact place and its fixed structure: it is the SINGLE authority for the location, the structure's shape, proportions, materials, colors, openings and every permanent site detail. Never copy its camera framing, time of day or lighting — the shot text is the authority for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/background_chain/seed_bg_forest_goat_pen_sel.png" } ], "critique_skipped": true, "ref_mode": "seed-bg만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only" }, "S110sh22": { "input_fingerprint": "a24c6c2260b05839", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, total lunar eclipse, blood moon.\n\nSHOT TEXT (authoritative, Korean): (온몸과 얼굴 전체에 먹물 같은 히브리 문자가 드러난 초자연적 형상으로 변한 김지후(강수리영))의 기괴한 전신.\n\nLOCATION (lock): The central sacrificial platform before the massive rock in the moonlit forest ritual clearing. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same giant rock altar, circular fires, black-clad crowd, smoke, and eclipse-red lighting. Exclude the elder as the central figure and show the transformed young woman bound at the altar with supernatural markings.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is immobilized upright at the central altar facing the ritual gathering, with both wrists and both ankles tightly secured by thick leather straps and her torso exposed before the bloodstained rock.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu's entire body and face remain covered in ink-black Hebrew-like markings, and her eyes remain supernaturally golden. She is immobilized upright at the central altar facing the ritual gathering, with both wrists and both ankles tightly secured by thick leather straps and her torso exposed before the bloodstained rock, unless the action explicitly breaks or releases the restraints.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (빙의된 초자연적 형상) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 황금빛 눈동자와 노출된 상체, 전신 문양 및 배경 요소(군중, 제단 목걸이)를 훌륭히 재현했으나 손목 결박 디테일이 다소 아쉽습니다." }, { "label": "B", "score": 5, "verdict_ko": "손목과 발목 결박은 표현되었으나 황금빛 눈동자가 없고, 상체가 옷으로 가려져 있으며 제단의 목걸이가 누락되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "군중과 제단 목걸이가 사라졌고, 바위에 결박되지 않았으며 상체 노출과 황금빛 눈동자 조건을 모두 무시했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S105sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영) (빙의된 초자연적 형상): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S105sh6" }, "lane_policy": "ab_select_bypass:prev" }, "S110sh27": { "input_fingerprint": "9ad6adad57298867", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, total lunar eclipse, blood moon.\n\nSHOT TEXT (authoritative, Korean): 고산들의 날카로운 창끝이 강영란(흰 머리 여자)의 가슴팍을 깊숙이 파고드는 끔찍한 찰나.\n\nLOCATION (lock): Directly before the bound sacrifice at the center of the firelit ritual ground. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same fire-ringed clearing, giant rock, worshippers, and blood-red eclipse light. Exclude the earlier orderly ritual pose and show the charging man driving the antler spear through the elder toward the transformed woman.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu retains the full-body black markings and golden eyes while still bound at the altar; Youngran's blood-soaked white robe remains unchanged as the antler-tipped spear pierces her chest toward Jihu.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강영란(흰 머리 여자) (60대 후반 한국인 여성, 온화하고 고운 얼굴, 가지런히 틀어 묶은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "C": 3, "A": 2, "B": 2 }, "selected": "C", "ranking": [ "C", "A", "B" ], "verdicts": [ { "label": "C", "score": 3, "verdict_ko": "창이 가슴을 파고드는 핵심 액션과 제단에 묶인 지후가 완전히 누락되었으나, 배경의 신도들과 바위의 핏자국 등 환경 디테일이 그나마 가장 많이 남아있습니다." }, { "label": "A", "score": 2, "verdict_ko": "가슴을 찌르는 창, 돌진하는 남자, 묶여 있어야 할 지후 등 프롬프트의 필수 요소가 모두 누락된 심각한 실패작입니다." }, { "label": "B", "score": 2, "verdict_ko": "핵심 인물과 액션이 모두 누락되었으며, 흰색이어야 할 한복에 프롬프트에 없는 녹색 깃이 추가되어 더욱 감점되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S110sh22_sel.png" }, { "label": "CHARACTER REFERENCE — 강영란(흰 머리 여자): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 사슴뿔 창: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S110sh22" }, "lane_policy": "ab_select_bypass:prev" }, "S110sh45": { "input_fingerprint": "e25311f91984be3b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night, total lunar eclipse, blood moon.\n\nSHOT TEXT (authoritative, Korean): 핏빛 보름달 아래, 비명과 피보라로 가득 찬 지옥 같은 살육의 도당을 무표정하게 내려다보는 (초자연적 형상으로 변한 김지후(강수리영))의 압도적인 전면.\n\nLOCATION (lock): The blood-soaked center of the forest ritual clearing beneath the eclipsed red moon, surrounded by fires and mass slaughter. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same monumental rock, circular bonfires, red moonlight, smoke, and enclosed forest clearing. Exclude the earlier orderly congregation and intact ritual formation, replacing them with the full chaotic massacre beneath the transformed woman's impassive gaze.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Jihu remains in her transformed state with golden eyes and black Hebrew-like markings, now with Sandeul's blood splashed across her face. Youngran's corpse lies where she fell with a pierced chest, Sandeul's headless body remains nearby, and the broken antler-tipped spear stays discarded at the ritual ground.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 김지후(강수리영) (빙의된 초자연적 형상) (20대 초반 한국인 여성, 예쁘장한 타원형 얼굴, 짙은색 중간 길이 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 5, "C": 3 }, "selected": "A", "ranking": [ "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "하얀 한복과 머리띠 등 캐릭터 레퍼런스의 복장을 가장 정확히 구현했으며, 지시대로 배경의 정돈된 무리를 배제하여 가장 우수하나 금빛 눈동자와 얼굴의 핏자국이 누락되었습니다." }, { "label": "B", "score": 5, "verdict_ko": "요구된 금빛 눈동자를 표현했으나 머리띠가 누락되었고, 프롬프트에서 명시적으로 배제하라고 지시한 '정돈된 무리'를 배경에 그대로 남겨두어 감점되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "주인공의 의상을 지시와 전혀 다른 검은색 한복으로 변경하는 심각한 오류를 범했으며, 명시된 허용 인물 외에 수많은 엑스트라를 배경에 추가하여 최하점입니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S110sh27_sel.png" }, { "label": "CHARACTER REFERENCE — 김지후(강수리영) (빙의된 초자연적 형상): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S110sh27" } }, "S1sh7": { "input_fingerprint": "d9c4f3fb1f10024b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 눈앞에 나타난 염전 소금창고 크기의 낡은 나무집을 향해 팔을 뻗고 상체를 기울인 채 한쪽 발이 공중에 떠 있는 달리기 자세의 십여 명의 조선인들.\n\nLOCATION (lock): The reed-covered approach directly in front of a small, weathered wooden storehouse. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dense winter reeds, stark monochrome look, and hard daylight. Exclude the firing soldiers and keep only the fleeing civilians and the newly revealed wooden shed.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 7, "B": 3, "C": 4 }, "selected": "A", "ranking": [ "A", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "나무집을 향해 달려가는 방향과 공간적 행동 관계를 유일하게 제대로 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "나무집을 등지고 달리는 방향 오류가 있으며, 팔을 뻗는 동작도 누락되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "요구된 달리기 자세와 동작은 잘 표현했으나, 나무집을 등지고 뛰는 치명적인 공간 배치 오류가 있습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S1sh5_sel.png" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S1sh5" } }, "S65sh4": { "input_fingerprint": "951e08f6c885f2e3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — primarily contemporary South Korea, with historical scenes set in 1670 Joseon and Japanese-occupied Korea roughly a century earlier; all people are Korean unless stated otherwise. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 붉은 불길이 솟구치는 식당 뒤편을 향해 달려가며 한쪽 발이 막 허공에 뜬 한국인 마을 청년들의 다급한 뒷모습.\n\nLOCATION (lock): The lane wrapping around the restaurant toward its rear yard at night. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark restaurant frontage, narrow village lane, guard positions, and moonlight. Exclude the static surveillance formation and show the youths suddenly running toward the fire behind the building.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The watchers abandon their posts and run toward the reported fire, leaving the restaurant temporarily unguarded.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 마을 청년#1 (한국인 젊은 남성, 평범한 얼굴, 짧은 짙은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "totals": { "A": 3, "B": 4, "C": 7 }, "selected": "C", "ranking": [ "C", "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "달려가는 인원이 7명으로 늘어나 연속성이 깨졌고, 식당 건물의 좌우 방향이 뒤집혀 장소 설정에 위배됩니다." }, { "label": "B", "score": 4, "verdict_ko": "이전 컷의 4명 중 3명만 등장해 인원수가 맞지 않으며, 우측에 없던 돌담이 추가되어 배경 연속성이 어긋납니다." }, { "label": "C", "score": 7, "verdict_ko": "4명의 청년이 화재를 향해 달리는 다급한 뒷모습과 허공에 뜬 발을 잘 묘사했으며, 이전 컷의 배경 구조와 인원수를 정확히 유지했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/475e5694-7c25-4f3f-8dec-910412a1761d/images/18f15068-4349-4818-82a9-3b9b173110e0/scene/recipe/S65sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 마을 청년#1: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique_skipped": true, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S65sh1" } } }