{ "S1sh1::variants": { "author_fp": "880e4cb7700f4dfa", "author": { "variants": [ { "approach_ko": "눈높이의 극도로 밀착된 정면 클로즈업으로 공포에 질린 얼굴과 수풀의 압박감을 강조한다.", "prompt_en": "Photograph the moment as an intensely tight, straight-on close-up at eye level, with the face nearly filling the frame and the surrounding branches pressing into the edges. Use a gently compressed lens feel and very shallow depth of field: hold the eyes and strained facial muscles in crisp focus while the nearest vegetation falls into soft, claustrophobic blur. Let the dim, cool dusk light filter unevenly through the brush, creating restrained natural shadow across the face." }, { "approach_ko": "더 넓고 깊은 정면 구도에서 수풀의 격자 속에 얼굴을 작게 고립시켜 도주로의 폐쇄감을 부각한다.", "prompt_en": "Pull back into a wider straight-on composition that makes the dense brush dominate, isolating the face within an irregular lattice of branches and vegetation. Favor a broader lens feel and layered depth, keeping both the foreground tangle and the face substantially legible so the confined escape route feels physically inescapable. Use the weak dusk illumination as soft ambient light, allowing the deeper forest to recede into natural darkness around the centered human presence." } ] }, "reused": false }, "S1sh1": { "input_fingerprint": "4e491bc1795dec0a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 어두컴컴한 숲속 수풀 사이, 잔뜩 겁에 질린 표정의 금월산 남성 피해자의 얼굴이 나뭇가지 밖으로 내밀어진 정면 구도.\n\nLOCATION (lock): A dense forest filled with overgrown brush, forming a confined escape route through thick vegetation. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 금월산 남성 피해자 (한국인 성인 남성, 짙은색 머리, 평범한 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 어두컴컴한 숲속 수풀 사이, 잔뜩 겁에 질린 표정의 금월산 남성 피해자의 얼굴이 나뭇가지 밖으로 내밀어진 정면 구도.\n\nLOCATION (lock): A dense forest filled with overgrown brush, forming a confined escape route through thick vegetation. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 금월산 남성 피해자 (한국인 성인 남성, 짙은색 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as an intensely tight, straight-on close-up at eye level, with the face nearly filling the frame and the surrounding branches pressing into the edges. Use a gently compressed lens feel and very shallow depth of field: hold the eyes and strained facial muscles in crisp focus while the nearest vegetation falls into soft, claustrophobic blur. Let the dim, cool dusk light filter unevenly through the brush, creating restrained natural shadow across the face.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 어두컴컴한 숲속 수풀 사이, 잔뜩 겁에 질린 표정의 금월산 남성 피해자의 얼굴이 나뭇가지 밖으로 내밀어진 정면 구도.\n\nLOCATION (lock): A dense forest filled with overgrown brush, forming a confined escape route through thick vegetation. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 금월산 남성 피해자 (한국인 성인 남성, 짙은색 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPull back into a wider straight-on composition that makes the dense brush dominate, isolating the face within an irregular lattice of branches and vegetation. Favor a broader lens feel and layered depth, keeping both the foreground tangle and the face substantially legible so the confined escape route feels physically inescapable. Use the weak dusk illumination as soft ambient light, allowing the deeper forest to recede into natural darkness around the centered human presence.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 5 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 정면 구도를 완벽히 구현했으며, 스토리보드에 제시된 양손의 포즈와 겁에 질린 표정을 충실하게 반영했습니다." }, { "label": "B", "score": 5, "verdict_ko": "전반적인 분위기는 잘 살렸으나, 시선이 측면을 향하고 있어 '정면 구도'라는 핵심 지시사항과 스토리보드의 디테일을 놓쳤습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S1sh1.png" }, { "label": "CHARACTER REFERENCE — 금월산 남성 피해자: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "화면 왼쪽의 주먹 쥔 손이 손가락 형태가 구별되지 않는 뭉개진 덩어리로 묘사되었습니다.", "fix_en": "Render the fist on the left with distinct, anatomically correct fingers." }, { "issue_ko": "화면 오른쪽 나뭇가지를 잡은 손의 검지손가락이 기형적으로 길고 전체적인 구조가 왜곡되었습니다.", "fix_en": "Correct the hand on the right to have normal human proportions and a realistic grip." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Render the fist on the left with distinct, anatomically correct fingers.\n- Correct the hand on the right to have normal human proportions and a realistic grip.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S1sh4::variants": { "author_fp": "79b185963c35930a", "author": { "variants": [ { "approach_ko": "낮은 측면 와이드 구도로 공중의 돌진 궤적과 바닥의 피해자를 한 프레임에 선명하게 대비한다.", "prompt_en": "Photograph the beat in a low, lateral wide composition, holding the fallen victim near the bottom of the frame while the airborne human figure cuts across the upper middle with both arms extended. Use layered brush and compressed clearings to make the escape route feel claustrophobic, with dusk light grazing through the vegetation and preserving believable human form and clothing detail within the figure’s dark silhouette." }, { "approach_ko": "피해자 가까이의 지면 시점에서 날아드는 인물을 강하게 단축해 위협과 충돌 직전의 충격을 강조한다.", "prompt_en": "Place the camera at ground level close to the fallen victim, looking along the confined route as the airborne figure surges toward the foreground with strongly foreshortened outstretched arms. Let the victim anchor the near edge while dense vegetation closes around the action; use the fading dusk behind and above the approaching figure to create a dark, readable human outline without losing realistic facial, hand, and wardrobe detail." } ] }, "reused": false }, "S1sh4": { "input_fingerprint": "8353ec11b069c27b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 바닥에 쓰러진 금월산 남성 피해자를 덮치기 위해 양팔을 뻗은 채 공중을 가로지르는 검은 인영의 실루엣.\n\nLOCATION (lock): A dense forest filled with overgrown brush, forming a confined escape route through thick vegetation. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 금월산 남성 피해자 (한국인 성인 남성, 짙은색 머리, 평범한 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 바닥에 쓰러진 금월산 남성 피해자를 덮치기 위해 양팔을 뻗은 채 공중을 가로지르는 검은 인영의 실루엣.\n\nLOCATION (lock): A dense forest filled with overgrown brush, forming a confined escape route through thick vegetation. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 금월산 남성 피해자 (한국인 성인 남성, 짙은색 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the beat in a low, lateral wide composition, holding the fallen victim near the bottom of the frame while the airborne human figure cuts across the upper middle with both arms extended. Use layered brush and compressed clearings to make the escape route feel claustrophobic, with dusk light grazing through the vegetation and preserving believable human form and clothing detail within the figure’s dark silhouette.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 바닥에 쓰러진 금월산 남성 피해자를 덮치기 위해 양팔을 뻗은 채 공중을 가로지르는 검은 인영의 실루엣.\n\nLOCATION (lock): A dense forest filled with overgrown brush, forming a confined escape route through thick vegetation. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 금월산 남성 피해자 (한국인 성인 남성, 짙은색 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera at ground level close to the fallen victim, looking along the confined route as the airborne figure surges toward the foreground with strongly foreshortened outstretched arms. Let the victim anchor the near edge while dense vegetation closes around the action; use the fading dusk behind and above the approaching figure to create a dark, readable human outline without losing realistic facial, hand, and wardrobe detail.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "스토리보드가 지정한 카메라 앵글과 인물의 배치(왼쪽에서 덮치는 인영, 오른쪽의 피해자)를 정확히 구현했으며 캐릭터 레퍼런스도 잘 반영했습니다." }, { "label": "B", "score": 3, "verdict_ko": "피해자의 등 뒤에서 촬영하는 앵글로 변경하여 스토리보드에 고정된 카메라 구도와 인물 배치 지시를 크게 위반했습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S1sh4.png" }, { "label": "CHARACTER REFERENCE — 금월산 남성 피해자: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "피해자가 바닥에 뻗은 왼손의 엄지손가락이 바깥쪽(왼쪽)에 위치하여 해부학적으로 오른손의 형태를 하고 있습니다.", "fix_en": "Redraw the victim's extended left hand so the thumb is on the inside edge, correctly depicting a left hand." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw the victim's extended left hand so the thumb is on the inside edge, correctly depicting a left hand.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S2sh3::variants": { "author_fp": "7c284b55309399b8", "author": { "variants": [ { "approach_ko": "바닥 가까운 넓은 시야와 깊은 초점으로 쓰러진 피해자와 고개를 든 남성을 같은 냉혹한 공간 안에 붙잡는다.", "prompt_en": "Photograph the moment from a low, ground-skimming viewpoint with a restrained wide-lens feel, using the alley floor as a strong foreground plane and the centered utility pole as the composition’s rigid vertical axis. Keep both figures clearly legible in deep focus, emphasizing the victim’s complete gravitational collapse against the man’s arrested upward posture. Let the faint streetlights create sparse edge highlights and broad pools of darkness, with realistic low-light texture and an unsensational, forensic stillness." }, { "approach_ko": "남성의 피 묻은 입가와 멈춘 표정을 중심으로 압축하고, 쓰러진 피해자는 얕은 초점 속 전경에 무겁게 남긴다.", "prompt_en": "Use a tighter, compressed perspective at roughly the man’s head height, framing his raised face as the immediate dramatic center while retaining the collapsed victim within the composition as a heavy foreground presence. Employ shallow, carefully placed focus so the blood at his mouth and the frozen tension of his expression are crisp while depth falls away naturally. Shape the faint streetlight into a narrow, uneven side illumination across his face, allowing the rest of the alley to remain subdued and claustrophobic." } ] }, "reused": false }, "S2sh3": { "input_fingerprint": "a1c2c85835139503", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, under faint streetlights.\n\nSHOT TEXT (authoritative, Korean): 효장동 여성 피해자의 몸이 바닥에 쓰러진 채 멈춰 있고, 입가에 피를 묻힌 한국인 남성이 고개를 든 정지 상태.\n\nLOCATION (lock): A narrow urban back alley centered on a utility pole, with faint streetlights providing sparse illumination. 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\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is fully collapsed and motionless on the ground, with her head, torso, arms, and legs unsupported and resting where they fell.\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 female victim is fully collapsed and motionless on the alley floor after collapsing, with her head, torso, arms, and legs unsupported and resting where they fell, with her neck bitten.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 효장동 여성 피해자 (한국인 성인 여성, 짙은색 머리, 평범한 얼굴); 효장동 한국인 남성 (포식자 활성 상태) (한국인 성인 남성, 짙은색 머리, 평범한 남성 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, under faint streetlights.\n\nSHOT TEXT (authoritative, Korean): 효장동 여성 피해자의 몸이 바닥에 쓰러진 채 멈춰 있고, 입가에 피를 묻힌 한국인 남성이 고개를 든 정지 상태.\n\nLOCATION (lock): A narrow urban back alley centered on a utility pole, with faint streetlights providing sparse illumination. 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\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is fully collapsed and motionless on the ground, with her head, torso, arms, and legs unsupported and resting where they fell.\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 female victim is fully collapsed and motionless on the alley floor after collapsing, with her head, torso, arms, and legs unsupported and resting where they fell, with her neck bitten.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 효장동 여성 피해자 (한국인 성인 여성, 짙은색 머리, 평범한 얼굴); 효장동 한국인 남성 (포식자 활성 상태) (한국인 성인 남성, 짙은색 머리, 평범한 남성 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment from a low, ground-skimming viewpoint with a restrained wide-lens feel, using the alley floor as a strong foreground plane and the centered utility pole as the composition’s rigid vertical axis. Keep both figures clearly legible in deep focus, emphasizing the victim’s complete gravitational collapse against the man’s arrested upward posture. Let the faint streetlights create sparse edge highlights and broad pools of darkness, with realistic low-light texture and an unsensational, forensic stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, under faint streetlights.\n\nSHOT TEXT (authoritative, Korean): 효장동 여성 피해자의 몸이 바닥에 쓰러진 채 멈춰 있고, 입가에 피를 묻힌 한국인 남성이 고개를 든 정지 상태.\n\nLOCATION (lock): A narrow urban back alley centered on a utility pole, with faint streetlights providing sparse illumination. 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\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is fully collapsed and motionless on the ground, with her head, torso, arms, and legs unsupported and resting where they fell.\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 female victim is fully collapsed and motionless on the alley floor after collapsing, with her head, torso, arms, and legs unsupported and resting where they fell, with her neck bitten.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 효장동 여성 피해자 (한국인 성인 여성, 짙은색 머리, 평범한 얼굴); 효장동 한국인 남성 (포식자 활성 상태) (한국인 성인 남성, 짙은색 머리, 평범한 남성 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a tighter, compressed perspective at roughly the man’s head height, framing his raised face as the immediate dramatic center while retaining the collapsed victim within the composition as a heavy foreground presence. Employ shallow, carefully placed focus so the blood at his mouth and the frozen tension of his expression are crisp while depth falls away naturally. Shape the faint streetlight into a narrow, uneven side illumination across his face, allowing the rest of the alley to remain subdued and claustrophobic.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 8, "B": 5 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "남성이 고개를 들고 있는 동작과 배경 중앙에 전봇대가 위치한 구도 등 텍스트의 핵심 지시사항을 충실히 반영했습니다." }, { "label": "B", "score": 5, "verdict_ko": "스토리보드의 여성 자세는 유사하게 구현했으나, 남성이 고개를 들고 있지 않으며 중앙 전봇대라는 배경 설정이 누락되었습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S2sh3.png" }, { "label": "CHARACTER REFERENCE — 효장동 한국인 남성 (포식자 활성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "화면 중앙의 전봇대 상단에 붙은 종이와 우측 건물의 파란색 간판에 텍스트가 포함되어 있습니다.", "fix_en": "Remove all text and numbers from the sticker on the utility pole and the blue sign on the right building." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text and numbers from the sticker on the utility pole and the blue sign on the right building.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S3sh2::variants": { "author_fp": "8f29dd9baf00b7b4", "author": { "variants": [ { "approach_ko": "눈높이의 압축된 미디엄 클로즈업으로 혜수의 굳은 표정과 무전기를 고립시키는 접근.", "prompt_en": "Frame an eye-level medium close-up with a restrained long-lens feel, keeping her face and radio hand crisp while the fixed crime-scene depth falls softly out of focus. Use the police tape in its established position as a strong visual divider, with balanced headroom and subdued natural daytime contrast emphasizing her arrested stillness." }, { "approach_ko": "가까운 사선 구도와 깊은 초점으로 혜수와 폴리스 라인의 긴장을 함께 살리는 접근.", "prompt_en": "Photograph the upper body from a close three-quarter angle with a moderately wide, immediate lens feel, preserving stronger depth through the fixed surroundings rather than isolating her. Let the police tape create a taut diagonal within its established placement, while directional daylight models her mature features and the radio hand; hold the composition slightly off-center to give her rigid pause unresolved tension." } ] }, "reused": false }, "S3sh2": { "input_fingerprint": "fd34942d00b37b4b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 무전기를 손에 쥔 채 폴리스 라인 밖에서 굳은 표정으로 멈춰 서 있는 혜수의 상체.\n\nLOCATION (lock): Two outdoor crime scenes—a wooded trail and an urban back alley—sealed with police tape while officers process and remove bodies. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 무전기를 손에 쥔 채 폴리스 라인 밖에서 굳은 표정으로 멈춰 서 있는 혜수의 상체.\n\nLOCATION (lock): Two outdoor crime scenes—a wooded trail and an urban back alley—sealed with police tape while officers process and remove bodies. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame an eye-level medium close-up with a restrained long-lens feel, keeping her face and radio hand crisp while the fixed crime-scene depth falls softly out of focus. Use the police tape in its established position as a strong visual divider, with balanced headroom and subdued natural daytime contrast emphasizing her arrested stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 무전기를 손에 쥔 채 폴리스 라인 밖에서 굳은 표정으로 멈춰 서 있는 혜수의 상체.\n\nLOCATION (lock): Two outdoor crime scenes—a wooded trail and an urban back alley—sealed with police tape while officers process and remove bodies. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the upper body from a close three-quarter angle with a moderately wide, immediate lens feel, preserving stronger depth through the fixed surroundings rather than isolating her. Let the police tape create a taut diagonal within its established placement, while directional daylight models her mature features and the radio hand; hold the composition slightly off-center to give her rigid pause unresolved tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 9, "A": 6 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "스토리보드에 지정된 구도, 인물의 시선 방향(우측), 그리고 배경의 건물 및 경찰차 위치를 완벽하게 구현하여 위치와 연출 기준을 가장 잘 충족합니다." }, { "label": "A", "score": 6, "verdict_ko": "캐릭터와 표정 연출은 우수하나, 스토리보드에 명시된 배경의 건물 구조물을 누락하고 인물의 시선 방향이 달라 연출 우선순위에서 감점되었습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S3sh2.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "폴리스 라인 테이프 표면에 글자가 포함되어 있어 'No text anywhere' 규칙을 위반했습니다.", "fix_en": "Remove all text from the yellow police tape, leaving it completely blank or with only plain stripes." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text from the yellow police tape, leaving it completely blank or with only plain stripes.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S4sh1::variants": { "author_fp": "ce690bb1a83a3b42", "author": { "variants": [ { "approach_ko": "주변 옥상과 옥탑방의 관계를 한눈에 읽히게 하는 넓고 정적인 정면 स्थाप립숏.", "prompt_en": "Photograph it as a wide, restrained establishing view from a neighboring rooftop-height vantage, with a natural perspective and level architectural lines. Let the rooftop dwelling anchor the composition while the surrounding rooftop structure remains clearly legible, using layered depth rather than dramatic distortion. Preserve the crisp, clear midday illumination and its honest surface detail, with a calm, observational stillness." }, { "approach_ko": "옥상 바닥 가까이에서 올려다보며 옥탑방의 낡은 외관과 하늘의 대비를 강조한 건축적 로우앵글.", "prompt_en": "Use a low rooftop-level viewpoint looking upward in a tighter architectural composition, giving the fixed structure a strong physical presence against the clear daytime sky while retaining enough of the surrounding rooftop context to establish its placement. Favor a moderately compressed lens feel, firm geometric framing, and a deliberate asymmetrical balance. Let the hard, lucid daytime light articulate the structure’s locked materials, openings, and weathered surfaces without stylization." } ] }, "reused": false }, "S4sh1": { "input_fingerprint": "2089f0eded314493", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 맑은 대낮의 인천 변두리, 낡은 다세대 빌라 옥상 위에 자리 잡은 옥탑방 외부 전경.\n\nLOCATION (lock): A rooftop dwelling built atop an old low-rise villa in an outlying urban neighborhood, viewed with the surrounding rooftop structure. 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\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 맑은 대낮의 인천 변두리, 낡은 다세대 빌라 옥상 위에 자리 잡은 옥탑방 외부 전경.\n\nLOCATION (lock): A rooftop dwelling built atop an old low-rise villa in an outlying urban neighborhood, viewed with the surrounding rooftop structure. 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\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph it as a wide, restrained establishing view from a neighboring rooftop-height vantage, with a natural perspective and level architectural lines. Let the rooftop dwelling anchor the composition while the surrounding rooftop structure remains clearly legible, using layered depth rather than dramatic distortion. Preserve the crisp, clear midday illumination and its honest surface detail, with a calm, observational stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 맑은 대낮의 인천 변두리, 낡은 다세대 빌라 옥상 위에 자리 잡은 옥탑방 외부 전경.\n\nLOCATION (lock): A rooftop dwelling built atop an old low-rise villa in an outlying urban neighborhood, viewed with the surrounding rooftop structure. 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\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a low rooftop-level viewpoint looking upward in a tighter architectural composition, giving the fixed structure a strong physical presence against the clear daytime sky while retaining enough of the surrounding rooftop context to establish its placement. Favor a moderately compressed lens feel, firm geometric framing, and a deliberate asymmetrical balance. Let the hard, lucid daytime light articulate the structure’s locked materials, openings, and weathered surfaces without stylization.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 6, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "위치 레퍼런스의 카메라 구도를 복사하지 말라는 지침을 준수하여, 옥상 위에서 바라본 옥탑방의 외부 전경을 텍스트에 맞게 새로운 앵글로 잘 담아냈습니다." }, { "label": "A", "score": 3, "verdict_ko": "위치 레퍼런스의 하이앵글 카메라 구도와 피사체 배치를 거의 그대로 복사하여 '구도를 복사하지 말라'는 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B02.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" } ], "critique": { "issues": [ { "issue_ko": "옥탑방 구조물의 외관(문의 색상 및 위치, 창문의 위치)이 'STRUCTURE LOOK' 기준 사진(왼쪽에 창문, 오른쪽에 녹색 문)을 따르지 않고 'LOCATION' 사진을 그대로 모방했습니다.", "fix_en": "Change the rooftop structure's facade to match the STRUCTURE LOOK photograph exactly, placing a window on the left and a green door on the right." }, { "issue_ko": "위성 안테나, 에어컨 실외기, 우측 은색 물탱크 표면에 텍스트가 생성되어 텍스트 금지 규칙을 위반했습니다.", "fix_en": "Remove all text from the satellite dish, the AC unit, and the silver water tank." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the rooftop structure's facade to match the STRUCTURE LOOK photograph exactly, placing a window on the left and a green door on the right.\n- Remove all text from the satellite dish, the AC unit, and the silver water tank.\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.", "ref_mode": "플레이트+seed만 (배경 전용)", "lane_policy": "ab_select_bypass:bg_only", "plate_select": { "candidates": { "A": "L03B01", "B": "L03B02", "C": "L03B03" }, "assigned": "L03B02", "choice": "Candidate B", "confident": true, "reason_ko": "모든 후보가 동일한 옥탑방 외부 전경(서브공간)을 보여주고 있으므로, '맑은 대낮'이라는 시간대 묘사에 가장 잘 부합하며 현재 할당되어 있는 Candidate B를 유지합니다.", "kept": "L03B02" } }, "S5sh1::variants": { "author_fp": "4b2ac591c523ccf1", "author": { "variants": [ { "approach_ko": "민숙의 옆얼굴과 굳은 시선을 밀착해 포착하고, 전경의 냄비와 김으로 불안감을 쌓는 친밀한 측면 구도.", "prompt_en": "Photograph her upper body in a tight side-profile close shot, holding her rigid expression and unwavering eyeline toward the television as the emotional center. Keep the steaming pot low in the near foreground, its white vapor rising through the narrow window light and partially veiling the cramped depth behind her; use shallow focus and restrained contrast for intimate, compressed tension." }, { "approach_ko": "TV 가까이에서 민숙을 정면으로 바라보며 부엌의 깊이와 냄비의 김을 함께 읽히게 하는 관찰자 시점 구도.", "prompt_en": "Place the camera close beside the television at seated eye height, looking back toward her so her upper body and fixed face register almost frontally while her gaze lands just off the camera. Use a moderately wide, deep composition that preserves the compact relationship between her, the stove, and the dining area, with the pot and its rising white steam clearly readable in a separate depth plane; let the narrow sunlight carve a firm directional strip through the modest interior." } ] }, "reused": false }, "S5sh1": { "input_fingerprint": "a2bf1da0f995d2af", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 냄비에서 하얀 김이 피어오르는 부엌, 앞치마를 두른 강민숙이 굳은 표정으로 TV 화면을 응시하는 상체.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk wears her apron and already bears the small circular mark on her inner wrist, which has been present since Suri-young was very young.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 냄비에서 하얀 김이 피어오르는 부엌, 앞치마를 두른 강민숙이 굳은 표정으로 TV 화면을 응시하는 상체.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk wears her apron and already bears the small circular mark on her inner wrist, which has been present since Suri-young was very young.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her upper body in a tight side-profile close shot, holding her rigid expression and unwavering eyeline toward the television as the emotional center. Keep the steaming pot low in the near foreground, its white vapor rising through the narrow window light and partially veiling the cramped depth behind her; use shallow focus and restrained contrast for intimate, compressed tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 냄비에서 하얀 김이 피어오르는 부엌, 앞치마를 두른 강민숙이 굳은 표정으로 TV 화면을 응시하는 상체.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk wears her apron and already bears the small circular mark on her inner wrist, which has been present since Suri-young was very young.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera close beside the television at seated eye height, looking back toward her so her upper body and fixed face register almost frontally while her gaze lands just off the camera. Use a moderately wide, deep composition that preserves the compact relationship between her, the stove, and the dining area, with the pot and its rising white steam clearly readable in a separate depth plane; let the narrow sunlight carve a firm directional strip through the modest interior.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 냄비에서 하얀 김이 피어오르는 부엌, 앞치마를 두른 강민숙이 굳은 표정으로 TV 화면을 응시하는 상체.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk wears her apron and already bears the small circular mark on her inner wrist, which has been present since Suri-young was very young.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her upper body in a tight side-profile close shot, holding her rigid expression and unwavering eyeline toward the television as the emotional center. Keep the steaming pot low in the near foreground, its white vapor rising through the narrow window light and partially veiling the cramped depth behind her; use shallow focus and restrained contrast for intimate, compressed tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 냄비에서 하얀 김이 피어오르는 부엌, 앞치마를 두른 강민숙이 굳은 표정으로 TV 화면을 응시하는 상체.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk wears her apron and already bears the small circular mark on her inner wrist, which has been present since Suri-young was very young.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera close beside the television at seated eye height, looking back toward her so her upper body and fixed face register almost frontally while her gaze lands just off the camera. Use a moderately wide, deep composition that preserves the compact relationship between her, the stove, and the dining area, with the pot and its rising white steam clearly readable in a separate depth plane; let the narrow sunlight carve a firm directional strip through the modest interior.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S5sh1.png" }, { "label": "CHARACTER REFERENCE — 강민숙: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S5sh1.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A", "D", "C" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "TV가 아닌 창밖을 향한 시선과 잘못된 위치(싱크대)에서 피어오르는 김이 주요 감점 요인입니다." }, { "label": "B", "score": 7, "verdict_ko": "TV를 응시하는 시선, 냄비의 김, 굳은 표정과 손목의 표식 등 모든 지시를 정확히 구현했습니다." }, { "label": "C", "score": 3, "verdict_ko": "시선이 TV를 향하지 않으며, 가스레인지와 싱크대의 공간 구조가 심하게 왜곡되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "TV 화면이 꺼져 있고 시선이 어긋나며, 지시되지 않은 숟가락을 들고 있어 감점되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A", "D", "C" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "TV가 아닌 창밖을 향한 시선과 잘못된 위치(싱크대)에서 피어오르는 김이 주요 감점 요인입니다." }, { "label": "B", "score": 7, "verdict_ko": "TV를 응시하는 시선, 냄비의 김, 굳은 표정과 손목의 표식 등 모든 지시를 정확히 구현했습니다." }, { "label": "C", "score": 3, "verdict_ko": "시선이 TV를 향하지 않으며, 가스레인지와 싱크대의 공간 구조가 심하게 왜곡되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "TV 화면이 꺼져 있고 시선이 어긋나며, 지시되지 않은 숟가락을 들고 있어 감점되었습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "TV 화면을 응시하는 핵심 행동을 유일하게 정확히 연출했으며, 손목의 표식과 상체 구도도 지시에 부합합니다." }, { "label": "C", "score": 5, "verdict_ko": "부엌 배경과 냄비의 김은 잘 표현되었으나, TV를 등지고 엉뚱한 곳을 보고 있어 핵심 행동 지시를 위반했습니다." }, { "label": "A", "score": 4, "verdict_ko": "TV를 보지 않고 정면을 향해 앉아 있으며, 불필요한 식기를 들고 있어 상황 묘사가 빗나갔습니다." }, { "label": "B", "score": 3, "verdict_ko": "부엌의 공간 구조가 레퍼런스와 다르게 왜곡되었고, TV를 응시하는 주요 행동도 누락되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "TV 화면을 응시하는 핵심 행동을 유일하게 정확히 연출했으며, 손목의 표식과 상체 구도도 지시에 부합합니다." }, { "label": "B", "score": 5, "verdict_ko": "부엌 배경과 냄비의 김은 잘 표현되었으나, TV를 등지고 엉뚱한 곳을 보고 있어 핵심 행동 지시를 위반했습니다." }, { "label": "D", "score": 4, "verdict_ko": "TV를 보지 않고 정면을 향해 앉아 있으며, 불필요한 식기를 들고 있어 상황 묘사가 빗나갔습니다." }, { "label": "C", "score": 3, "verdict_ko": "부엌의 공간 구조가 레퍼런스와 다르게 왜곡되었고, TV를 응시하는 주요 행동도 누락되었습니다." } ] }, "combined": { "totals": { "A": 11, "B": 12, "C": 6, "D": 8 }, "ranking": [ "B", "A", "D", "C" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 12, "C": 6, "D": 8 }, "selected": "B", "ranking": [ "B", "A", "D", "C" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "TV가 아닌 창밖을 향한 시선과 잘못된 위치(싱크대)에서 피어오르는 김이 주요 감점 요인입니다." }, { "label": "B", "score": 7, "verdict_ko": "TV를 응시하는 시선, 냄비의 김, 굳은 표정과 손목의 표식 등 모든 지시를 정확히 구현했습니다." }, { "label": "C", "score": 3, "verdict_ko": "시선이 TV를 향하지 않으며, 가스레인지와 싱크대의 공간 구조가 심하게 왜곡되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "TV 화면이 꺼져 있고 시선이 어긋나며, 지시되지 않은 숟가락을 들고 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "모든 텍스트를 금지하는 지시사항이 있으나, 우측 TV 화면에 뉴스 자막 형태의 텍스트가 포함되어 있습니다.", "fix_en": "Remove all text and captions from the television screen." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text and captions from the television screen.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·B=변형1)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B01", "choice": "Candidate A", "confident": true, "reason_ko": "지문에서 명시한 '냄비에서 하얀 김이 피어오르는 부엌'과 'TV 화면'이 모두 한 프레임 안에 가장 잘 구현된 공간이 후보 A입니다.", "kept": "L04B01" } }, "S5sh6::variants": { "author_fp": "68e61a7c74082ce2", "author": { "variants": [ { "approach_ko": "어깨 높이의 밀착 클로즈업으로 수리영의 감긴 눈과 강민숙의 어깨에 닿은 얼굴을 섬세하게 포착한다.", "prompt_en": "Photograph the moment in an intimate shoulder-height close-up with a gently compressed lens feel. Frame Suriyoung’s closed eyes and the physical contact of her nose against Minsuk’s shoulder as the visual center, letting Minsuk’s apron and partial profile anchor the foreground. Use shallow depth of field and narrow window light grazing their faces and fabric, preserving the absolute stillness of the pose." }, { "approach_ko": "방 안을 함께 담는 절제된 중거리 정면 구도로 두 사람의 멈춘 자세와 비좁은 생활 공간의 정적을 강조한다.", "prompt_en": "Use a restrained medium-wide view from across the compact room, at natural standing height, holding both women within the closely packed domestic interior. Arrange the window, kitchenette, dining table, and bedroom thresholds as quiet depth layers around their motionless embrace, with clean negative space emphasizing the pause. Let the narrow daylight remain directional and localized, while the rest of the room falls into soft, natural interior shadow with broad depth of field." } ] }, "reused": false }, "S5sh6": { "input_fingerprint": "abf8def4b1fcefd4", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 강민숙의 어깨에 코를 묻은 채 눈을 감고 멈춘 상태.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk remains in her apron and retains the longstanding small circular mark on her inner wrist.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 얼굴); 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 강민숙의 어깨에 코를 묻은 채 눈을 감고 멈춘 상태.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk remains in her apron and retains the longstanding small circular mark on her inner wrist.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 얼굴); 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in an intimate shoulder-height close-up with a gently compressed lens feel. Frame Suriyoung’s closed eyes and the physical contact of her nose against Minsuk’s shoulder as the visual center, letting Minsuk’s apron and partial profile anchor the foreground. Use shallow depth of field and narrow window light grazing their faces and fabric, preserving the absolute stillness of the pose.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 강민숙의 어깨에 코를 묻은 채 눈을 감고 멈춘 상태.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk remains in her apron and retains the longstanding small circular mark on her inner wrist.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 얼굴); 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained medium-wide view from across the compact room, at natural standing height, holding both women within the closely packed domestic interior. Arrange the window, kitchenette, dining table, and bedroom thresholds as quiet depth layers around their motionless embrace, with clean negative space emphasizing the pause. Let the narrow daylight remain directional and localized, while the rest of the room falls into soft, natural interior shadow with broad depth of field.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 강민숙의 어깨에 코를 묻은 채 눈을 감고 멈춘 상태.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk remains in her apron and retains the longstanding small circular mark on her inner wrist.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 얼굴); 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in an intimate shoulder-height close-up with a gently compressed lens feel. Frame Suriyoung’s closed eyes and the physical contact of her nose against Minsuk’s shoulder as the visual center, letting Minsuk’s apron and partial profile anchor the foreground. Use shallow depth of field and narrow window light grazing their faces and fabric, preserving the absolute stillness of the pose.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 강민숙의 어깨에 코를 묻은 채 눈을 감고 멈춘 상태.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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): Minsuk remains in her apron and retains the longstanding small circular mark on her inner wrist.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 얼굴); 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained medium-wide view from across the compact room, at natural standing height, holding both women within the closely packed domestic interior. Arrange the window, kitchenette, dining table, and bedroom thresholds as quiet depth layers around their motionless embrace, with clean negative space emphasizing the pause. Let the narrow daylight remain directional and localized, while the rest of the room falls into soft, natural interior shadow with broad depth of field.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S5sh6.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S5sh6.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "C", "ranking": [ "C", "D", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "어깨에 코를 묻고 눈을 감은 수리영의 모습을 밀착된 구도로 완벽히 담아냈으며 어떤 규정도 위반하지 않았습니다." }, { "label": "D", "score": 5, "verdict_ko": "풀샷 구도라 지시된 세밀한 행동 묘사가 약하며, 강민숙의 손목에 있어야 할 원형 마크가 수리영의 팔에 있어 감점되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "강민숙의 손목에 위치해야 할 원형 마크가 수리영의 팔에 두 개나 중복되어 나타나는 심각한 오류가 있습니다." }, { "label": "B", "score": 3, "verdict_ko": "인물들의 팔이 불가능한 형태로 얽혀 해부학적 오류를 보이며, 마크가 잘못된 인물의 옷소매에 위치해 있습니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "D", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "어깨에 코를 묻고 눈을 감은 수리영의 모습을 밀착된 구도로 완벽히 담아냈으며 어떤 규정도 위반하지 않았습니다." }, { "label": "D", "score": 5, "verdict_ko": "풀샷 구도라 지시된 세밀한 행동 묘사가 약하며, 강민숙의 손목에 있어야 할 원형 마크가 수리영의 팔에 있어 감점되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "강민숙의 손목에 위치해야 할 원형 마크가 수리영의 팔에 두 개나 중복되어 나타나는 심각한 오류가 있습니다." }, { "label": "B", "score": 3, "verdict_ko": "인물들의 팔이 불가능한 형태로 얽혀 해부학적 오류를 보이며, 마크가 잘못된 인물의 옷소매에 위치해 있습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A", "C", "D" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "수리영이 어깨에 코를 묻고 눈을 감은 텍스트의 핵심 동작을 적절한 앵글로 정확히 포착했으며, 배경의 장소도 일치함." }, { "label": "A", "score": 5, "verdict_ko": "지정된 동작을 표현했으나 프레임이 불필요하게 넓고, 민숙이 아닌 수리영의 팔에 원형 자국이 잘못 배치됨." }, { "label": "C", "score": 4, "verdict_ko": "회색 긴소매를 입은 팔의 위치와 구조가 해부학적으로 어색하며, 원형 자국 역시 잘못된 인물에게 그려짐." }, { "label": "D", "score": 3, "verdict_ko": "원형 자국이 있는 손과 팔이 여러 개로 중복되어 나타나는 물리적으로 불가능한 해부학적 오류(Hard Violation)가 있음." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "D", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "수리영이 어깨에 코를 묻고 눈을 감은 텍스트의 핵심 동작을 적절한 앵글로 정확히 포착했으며, 배경의 장소도 일치함." }, { "label": "D", "score": 5, "verdict_ko": "지정된 동작을 표현했으나 프레임이 불필요하게 넓고, 민숙이 아닌 수리영의 팔에 원형 자국이 잘못 배치됨." }, { "label": "B", "score": 4, "verdict_ko": "회색 긴소매를 입은 팔의 위치와 구조가 해부학적으로 어색하며, 원형 자국 역시 잘못된 인물에게 그려짐." }, { "label": "A", "score": 3, "verdict_ko": "원형 자국이 있는 손과 팔이 여러 개로 중복되어 나타나는 물리적으로 불가능한 해부학적 오류(Hard Violation)가 있음." } ] }, "combined": { "totals": { "A": 7, "B": 7, "C": 14, "D": 10 }, "ranking": [ "C", "D", "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 7, "B": 7, "C": 14, "D": 10 }, "selected": "C", "ranking": [ "C", "D", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "어깨에 코를 묻고 눈을 감은 수리영의 모습을 밀착된 구도로 완벽히 담아냈으며 어떤 규정도 위반하지 않았습니다." }, { "label": "D", "score": 5, "verdict_ko": "풀샷 구도라 지시된 세밀한 행동 묘사가 약하며, 강민숙의 손목에 있어야 할 원형 마크가 수리영의 팔에 있어 감점되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "강민숙의 손목에 위치해야 할 원형 마크가 수리영의 팔에 두 개나 중복되어 나타나는 심각한 오류가 있습니다." }, { "label": "B", "score": 3, "verdict_ko": "인물들의 팔이 불가능한 형태로 얽혀 해부학적 오류를 보이며, 마크가 잘못된 인물의 옷소매에 위치해 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영의 눈이 완전히 감기지 않고 아래를 향해 살짝 떠 있습니다.", "fix_en": "Fully close Suriyoung's eyes so that her eyelids are completely shut and no part of the eye is visible." }, { "issue_ko": "강민숙이 레퍼런스의 베이지색 티셔츠와 데님 앞치마를 입지 않고, 전체가 데님 소재인 긴팔 셔츠를 입고 있습니다.", "fix_en": "Replace Minsuk's full denim shirt with the beige t-shirt and light blue denim apron from her reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Fully close Suriyoung's eyes so that her eyelids are completely shut and no part of the eye is visible.\n- Replace Minsuk's full denim shirt with the beige t-shirt and light blue denim apron from her reference.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+엔티티 (4택1: 무콘티 승·C=변형0)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B02", "choice": "Candidate B", "confident": false, "reason_ko": "지문에 특정 공간(주방, 거실 등)을 지시하는 단서가 없어, 기존에 배정된 서브공간을 유지함.", "kept": "L04B02" } }, "S5sh7::variants": { "author_fp": "b59aeb70c146a1a1", "author": { "variants": [ { "approach_ko": "손목 안쪽을 정면에 가깝게 포착한 극단적 클로즈업으로 표식을 선명하고 객관적인 증거처럼 강조한다.", "prompt_en": "Photograph the inner wrist in an extreme close-up from a nearly perpendicular angle, with the small circular mark held precisely at the center of the frame. Use a restrained, shallow-focus lens feel: the mark and immediate skin texture are crisply resolved while the unchanged clothing and nearby wall material fall softly out of focus. Let the narrow window light model the wrist gently and naturally, creating an objective, evidentiary insert." }, { "approach_ko": "팔을 대각선으로 가로지르는 비스듬한 근접 시점과 스치는 창빛으로 표식의 발견 순간을 긴장감 있게 만든다.", "prompt_en": "Take a close, oblique viewpoint along the unchanged arm so the wrist runs diagonally through the composition and the circular mark remains centered. Use a more dimensional depth arrangement, with a trace of unchanged clothing near the foreground and the modest room surface receding behind the wrist. Allow the narrow sunlight to rake across the inner wrist, revealing the mark through gentle tonal contrast and giving the discovery an intimate, tense immediacy." } ] }, "reused": false }, "S5sh7": { "input_fingerprint": "9abc9a4b2c2f7c64", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 강민숙의 손목 안쪽에 새겨진 뚜렷한 작은 원형 표식이 화면 중앙에 보이는 시점 쇼트.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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 modest rooftop-room interior lighting, nearby wall materials, and the woman's unchanged clothing and arm position. Preserve the close physical proximity established between the mother and daughter. Exclude both faces and most of their bodies, as well as unrelated furniture, so the frame isolates the marked inner wrist.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 longstanding small circular mark remains clearly visible on the inside of Minsuk’s wrist.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 강민숙의 손목 안쪽에 새겨진 뚜렷한 작은 원형 표식이 화면 중앙에 보이는 시점 쇼트.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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 modest rooftop-room interior lighting, nearby wall materials, and the woman's unchanged clothing and arm position. Preserve the close physical proximity established between the mother and daughter. Exclude both faces and most of their bodies, as well as unrelated furniture, so the frame isolates the marked inner wrist.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 longstanding small circular mark remains clearly visible on the inside of Minsuk’s wrist.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the inner wrist in an extreme close-up from a nearly perpendicular angle, with the small circular mark held precisely at the center of the frame. Use a restrained, shallow-focus lens feel: the mark and immediate skin texture are crisply resolved while the unchanged clothing and nearby wall material fall softly out of focus. Let the narrow window light model the wrist gently and naturally, creating an objective, evidentiary insert.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 강민숙의 손목 안쪽에 새겨진 뚜렷한 작은 원형 표식이 화면 중앙에 보이는 시점 쇼트.\n\nLOCATION (lock): A compact rooftop apartment with a small living-room window, a kitchenette sink and stove, a dining table, and adjoining bedrooms. The interior is modest and closely packed. 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 modest rooftop-room interior lighting, nearby wall materials, and the woman's unchanged clothing and arm position. Preserve the close physical proximity established between the mother and daughter. Exclude both faces and most of their bodies, as well as unrelated furniture, so the frame isolates the marked inner wrist.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 longstanding small circular mark remains clearly visible on the inside of Minsuk’s wrist.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTake a close, oblique viewpoint along the unchanged arm so the wrist runs diagonally through the composition and the circular mark remains centered. Use a more dimensional depth arrangement, with a trace of unchanged clothing near the foreground and the modest room surface receding behind the wrist. Allow the narrow sunlight to rake across the inner wrist, revealing the mark through gentle tonal contrast and giving the discovery an intimate, tense immediacy.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 7, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "이전 샷의 팔 위치와 모녀간의 밀착 상태가 완전히 유지되지는 않았으나, 지정된 장소와 좁은 햇빛 조명을 정확히 구현하며 손목 안쪽의 원형 표식을 지시문대로 화면 중앙에 잘 강조했습니다." }, { "label": "A", "score": 3, "verdict_ko": "포커스 아웃된 배경의 인물이 두 사람의 의상(회색 티셔츠와 데님 앞치마)이 하나로 병합된 기괴한 형태로 나타나 신체 및 복장 구조상 치명적인 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/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/scene/recipe/S5sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "이전 샷의 포옹 자세가 풀린 채 팔이 허공으로 들려 있으며, 밀착해 있어야 할 딸(어두운 티셔츠)의 모습이 프레임에서 완전히 누락되었습니다.", "fix_en": "Reposition the arm horizontally to maintain the locked hugging pose, and place a portion of the daughter's dark t-shirt directly against the arm to restore the established physical contact." }, { "issue_ko": "손의 엄지손가락(오른쪽)이 비정상적으로 길고 마디가 많아 일반적인 손가락처럼 변형되어 있습니다.", "fix_en": "Correct the thumb's anatomy to have normal human proportions, thickness, and joints." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Reposition the arm horizontally to maintain the locked hugging pose, and place a portion of the daughter's dark t-shirt directly against the arm to restore the established physical contact.\n- Correct the thumb's anatomy to have normal human proportions, thickness, and joints.\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.", "ref_mode": "prev+엔티티" }, "S6sh1::variants": { "author_fp": "6bc8b00705ac6452", "author": { "variants": [ { "approach_ko": "허리 높이의 평행 트래킹 시점으로 인물과 자전거의 측면 동작을 또렷하게 분리해 담는 역동적 전신 숏.", "prompt_en": "Photograph the action from a parallel, waist-height tracking viewpoint, holding the complete side-on figure and bicycle cleanly within the frame. Use a moderately compressed lens feel to keep her expression readable while arranging the road and sea as distinct depth layers behind her. Crisp natural daylight, restrained highlights, and a fast, sharply resolved mid-action instant emphasize the force of the downstroke and the opposing sweep of her hair." }, { "approach_ko": "노면 가까운 낮은 시점과 넓은 원근감으로 페달의 추진력과 해안 공간을 크게 느끼게 하는 측면 전신 숏.", "prompt_en": "Take a low roadside viewpoint with a broad, immersive lens feel, preserving the full side profile while giving the bicycle and forceful pedal stroke strong foreground presence. Compose with generous coastal space around her so the open setting amplifies the buoyant momentum of the moment, while keeping her face clearly visible. Bright clear-day illumination, natural contrast, and slight environmental motion rendering create speed without softening her expression or body position." } ] }, "reused": false }, "S6sh1": { "input_fingerprint": "5d33203eeea7340a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 맑은 낮 해안가 도로, 이어폰을 낀 수리영이 밝은 표정으로 자전거 안장에 앉아 한쪽 페달을 아래로 힘차게 밟고 있으며 주행 반대 방향으로 머리카락이 흩날리는 mid-action 측면 전신.\n\nLOCATION (lock): A coastal road running directly alongside the sea, with enough roadside space for bicycle travel. 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): Suri-young continues wearing her earphones while riding the bicycle along the coast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 맑은 낮 해안가 도로, 이어폰을 낀 수리영이 밝은 표정으로 자전거 안장에 앉아 한쪽 페달을 아래로 힘차게 밟고 있으며 주행 반대 방향으로 머리카락이 흩날리는 mid-action 측면 전신.\n\nLOCATION (lock): A coastal road running directly alongside the sea, with enough roadside space for bicycle travel. 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): Suri-young continues wearing her earphones while riding the bicycle along the coast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the action from a parallel, waist-height tracking viewpoint, holding the complete side-on figure and bicycle cleanly within the frame. Use a moderately compressed lens feel to keep her expression readable while arranging the road and sea as distinct depth layers behind her. Crisp natural daylight, restrained highlights, and a fast, sharply resolved mid-action instant emphasize the force of the downstroke and the opposing sweep of her hair.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 맑은 낮 해안가 도로, 이어폰을 낀 수리영이 밝은 표정으로 자전거 안장에 앉아 한쪽 페달을 아래로 힘차게 밟고 있으며 주행 반대 방향으로 머리카락이 흩날리는 mid-action 측면 전신.\n\nLOCATION (lock): A coastal road running directly alongside the sea, with enough roadside space for bicycle travel. 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): Suri-young continues wearing her earphones while riding the bicycle along the coast.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTake a low roadside viewpoint with a broad, immersive lens feel, preserving the full side profile while giving the bicycle and forceful pedal stroke strong foreground presence. Compose with generous coastal space around her so the open setting amplifies the buoyant momentum of the moment, while keeping her face clearly visible. Bright clear-day illumination, natural contrast, and slight environmental motion rendering create speed without softening her expression or body position.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 5 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "스토리보드의 앵글과 배경 구조를 매우 정확하게 재현했으며, 페달을 밟는 역동적인 자세와 흩날리는 머리카락이 자연스럽게 묘사되었습니다." }, { "label": "B", "score": 5, "verdict_ko": "요구된 프레이밍과 포즈는 대체로 준수하였으나, 자전거 바퀴의 스포크와 크랭크 주변 구조가 기하학적으로 어색하게 묘사되었습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S6sh1.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물이 자전거 안장에서 엉덩이를 떼고 일어선 자세로, '자전거 안장에 앉아'라는 프롬프트 지시를 위반했습니다.", "fix_en": "Lower the character's body so she is seated firmly on the bicycle saddle." }, { "issue_ko": "캐릭터 레퍼런스에 있는 파란색 집업 재킷을 입지 않고 스토리보드의 임시 복장(회색 셔츠)을 그대로 모방했습니다.", "fix_en": "Clothe the character in the blue zip-up jacket shown in the character reference." }, { "issue_ko": "캐릭터 레퍼런스 이미지에서 착용하고 있는 검은색 자전거 헬멧이 누락되었습니다.", "fix_en": "Add the black bicycle helmet from the character reference onto the character's head." }, { "issue_ko": "캐릭터 레퍼런스에 등장하는 가슴을 가로지르는 올리브색 크로스백(패니팩)이 누락되었습니다.", "fix_en": "Draw the olive green fanny pack worn diagonally across the character's chest." }, { "issue_ko": "오른쪽 페달이 맨 아래(6시 방향)에 있으므로 왼쪽 다리는 굽혀져 위로 올라와야 하나, 자전거 반대편에 왼쪽 다리가 전혀 보이지 않습니다.", "fix_en": "Render the character's left leg bent upwards on the opposite side of the bicycle frame." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Lower the character's body so she is seated firmly on the bicycle saddle.\n- Clothe the character in the blue zip-up jacket shown in the character reference.\n- Add the black bicycle helmet from the character reference onto the character's head.\n- Draw the olive green fanny pack worn diagonally across the character's chest.\n- Render the character's left leg bent upwards on the opposite side of the bicycle frame.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S7sh2::variants": { "author_fp": "38719350fce5e3eb", "author": { "variants": [ { "approach_ko": "개의 눈높이에서 철창 너머 수리영을 정면 대칭으로 압축해 무표정한 응시를 강조한다.", "prompt_en": "Photograph her in a tight, frontal upper-body portrait from the dog’s eye level, with the camera positioned across the kennel bars. Use a compressed lens feel and near-symmetrical framing, keeping her unwavering gaze and restrained facial muscles critically sharp while the bars fall softly out of focus in the immediate foreground. Shape the available daytime light gently across her face, preserving a quiet, clinical stillness." }, { "approach_ko": "조금 더 넓은 정면 구도와 깊은 초점으로 수리영, 철창, 보호소 공간의 긴장을 함께 담는다.", "prompt_en": "Frame a wider frontal medium shot that retains her full upper body and gives the kennel bars a strong geometric presence between camera and subject. Use a natural perspective with deeper focus, arranging the shelter space in receding layers around her while her fixed gaze remains the compositional anchor. Let the practical daytime illumination feel harder and more observational, emphasizing physical distance and emotional restraint." } ] }, "reused": false }, "S7sh2": { "input_fingerprint": "398b57753e3e27e7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 밖에서 무표정한 얼굴로 개의 눈을 뚫어지게 응시하는 수리영의 정면 상체.\n\nLOCATION (lock): A dog shelter with enclosed kennels for multiple dogs and practical areas for veterinary injections, grooming, treatment, and washing. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 밖에서 무표정한 얼굴로 개의 눈을 뚫어지게 응시하는 수리영의 정면 상체.\n\nLOCATION (lock): A dog shelter with enclosed kennels for multiple dogs and practical areas for veterinary injections, grooming, treatment, and washing. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a tight, frontal upper-body portrait from the dog’s eye level, with the camera positioned across the kennel bars. Use a compressed lens feel and near-symmetrical framing, keeping her unwavering gaze and restrained facial muscles critically sharp while the bars fall softly out of focus in the immediate foreground. Shape the available daytime light gently across her face, preserving a quiet, clinical stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 밖에서 무표정한 얼굴로 개의 눈을 뚫어지게 응시하는 수리영의 정면 상체.\n\nLOCATION (lock): A dog shelter with enclosed kennels for multiple dogs and practical areas for veterinary injections, grooming, treatment, and washing. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a wider frontal medium shot that retains her full upper body and gives the kennel bars a strong geometric presence between camera and subject. Use a natural perspective with deeper focus, arranging the shelter space in receding layers around her while her fixed gaze remains the compositional anchor. Let the practical daytime illumination feel harder and more observational, emphasizing physical distance and emotional restraint.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 밖에서 무표정한 얼굴로 개의 눈을 뚫어지게 응시하는 수리영의 정면 상체.\n\nLOCATION (lock): A dog shelter with enclosed kennels for multiple dogs and practical areas for veterinary injections, grooming, treatment, and washing. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a tight, frontal upper-body portrait from the dog’s eye level, with the camera positioned across the kennel bars. Use a compressed lens feel and near-symmetrical framing, keeping her unwavering gaze and restrained facial muscles critically sharp while the bars fall softly out of focus in the immediate foreground. Shape the available daytime light gently across her face, preserving a quiet, clinical stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 밖에서 무표정한 얼굴로 개의 눈을 뚫어지게 응시하는 수리영의 정면 상체.\n\nLOCATION (lock): A dog shelter with enclosed kennels for multiple dogs and practical areas for veterinary injections, grooming, treatment, and washing. 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\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a wider frontal medium shot that retains her full upper body and gives the kennel bars a strong geometric presence between camera and subject. Use a natural perspective with deeper focus, arranging the shelter space in receding layers around her while her fixed gaze remains the compositional anchor. Let the practical daytime illumination feel harder and more observational, emphasizing physical distance and emotional restraint.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_animal_shelter_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S7sh2.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_animal_shelter_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S7sh2.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_animal_shelter_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_animal_shelter_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "C", "ranking": [ "C", "A", "D", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 정면 상체 구도, 개의 눈을 응시하는 시선, 기준 의상(파란 재킷)을 가장 정확하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "상체 구도는 맞으나 응시해야 할 개가 화면에 없고 기준 의상이 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "'정면 상체' 구도 지시를 어긴 전신 풀샷이며, 개와 시선을 교환하지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "실사 지침을 위반하고 2D 만화 스타일의 개가 합성되어 있어 하드 위반(콜라주)에 해당합니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "A", "D", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 정면 상체 구도, 개의 눈을 응시하는 시선, 기준 의상(파란 재킷)을 가장 정확하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "상체 구도는 맞으나 응시해야 할 개가 화면에 없고 기준 의상이 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "'정면 상체' 구도 지시를 어긴 전신 풀샷이며, 개와 시선을 교환하지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "실사 지침을 위반하고 2D 만화 스타일의 개가 합성되어 있어 하드 위반(콜라주)에 해당합니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "D", "A", "C" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "개와 인물은 포함되었으나, 상체 샷이 아닌 전신 와이드 샷으로 촬영되어 프레이밍 우선순위를 위반했습니다." }, { "label": "B", "score": 7, "verdict_ko": "지시된 정면 상체 프레이밍을 완벽히 따랐으며, 전경에 개의 뒷모습을 배치해 시선 교환과 상호작용을 훌륭히 구현했습니다." }, { "label": "C", "score": 3, "verdict_ko": "화면 전경의 개가 실사가 아닌 2D 애니메이션처럼 그려져 있어 합성(Collage) 하드 위반에 해당합니다." }, { "label": "D", "score": 5, "verdict_ko": "정면 상체 프레이밍은 준수했으나, 응시해야 할 개가 프레임 안에 없어 핵심적인 상호작용 지시가 누락되었습니다." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "A", "D", "B" ], "verdicts": [ { "label": "D", "score": 4, "verdict_ko": "개와 인물은 포함되었으나, 상체 샷이 아닌 전신 와이드 샷으로 촬영되어 프레이밍 우선순위를 위반했습니다." }, { "label": "C", "score": 7, "verdict_ko": "지시된 정면 상체 프레이밍을 완벽히 따랐으며, 전경에 개의 뒷모습을 배치해 시선 교환과 상호작용을 훌륭히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "화면 전경의 개가 실사가 아닌 2D 애니메이션처럼 그려져 있어 합성(Collage) 하드 위반에 해당합니다." }, { "label": "A", "score": 5, "verdict_ko": "정면 상체 프레이밍은 준수했으나, 응시해야 할 개가 프레임 안에 없어 핵심적인 상호작용 지시가 누락되었습니다." } ] }, "combined": { "totals": { "A": 10, "B": 6, "C": 14, "D": 8 }, "ranking": [ "C", "A", "D", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 10, "B": 6, "C": 14, "D": 8 }, "selected": "C", "ranking": [ "C", "A", "D", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 정면 상체 구도, 개의 눈을 응시하는 시선, 기준 의상(파란 재킷)을 가장 정확하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "상체 구도는 맞으나 응시해야 할 개가 화면에 없고 기준 의상이 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "'정면 상체' 구도 지시를 어긴 전신 풀샷이며, 개와 시선을 교환하지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "실사 지침을 위반하고 2D 만화 스타일의 개가 합성되어 있어 하드 위반(콜라주)에 해당합니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_animal_shelter_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물이 숏 텍스트에 명시된 '철창 밖에서'가 아니라, 레퍼런스의 콘크리트 벽으로 둘러싸인 견사 내부에 서 있습니다.", "fix_en": "Move the character so she is standing outside the kennel structure, not inside the concrete run." }, { "issue_ko": "캐릭터 레퍼런스에 포함된 검은색 자전거 헬멧을 착용하지 않았습니다.", "fix_en": "Add the black bicycle helmet from the reference image onto the character's head." }, { "issue_ko": "프레임 내에 보여야 할 초록색 힙색과 어깨를 가로지르는 스트랩이 누락되었습니다.", "fix_en": "Add the green fanny pack and its cross-body strap over the character's jacket." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the character so she is standing outside the kennel structure, not inside the concrete run.\n- Add the black bicycle helmet from the reference image onto the character's head.\n- Add the green fanny pack and its cross-body strap over the character's jacket.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "seed-bg+엔티티 (복잡구조물 4택1: 무콘티 승·C=변형0)", "lane_policy": "ab_select_ready" }, "S7sh5::variants": { "author_fp": "e3f38f43dfc497c5", "author": { "variants": [ { "approach_ko": "눈높이 측면 미디엄 클로즈업으로 수리영의 멈춘 표정과 우리 행렬의 깊이를 함께 포착한다.", "prompt_en": "Photograph her upper body in an intimate eye-level profile medium close-up, holding her face, parted lips, and phone hand in crisp focus while her fixed gaze leads across the frame. Let the metal kennel bars and contained dogs recede behind her in shallow, layered depth, with soft directional daylight shaping her young face and the worn metal surfaces. Keep the composition quiet and suspended, emphasizing the unanswered-call stillness." }, { "approach_ko": "우리 안쪽의 낮은 시점에서 전경 철창 너머 수리영을 바라보며 갇힘과 거리감을 강조한다.", "prompt_en": "Place the camera at a low kennel-side viewpoint, looking through prominent out-of-focus metal bars toward her upper body beyond them. Use a restrained wider lens feel and deeper spatial layering so the enclosure dominates the foreground while her halted posture and slightly open mouth remain clearly readable in the daylight. Compose her between the bars, making the physical separation from the dogs the central visual tension." } ] }, "reused": false }, "S7sh5": { "input_fingerprint": "6d15a54b21e93b6d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 갇힌 개들을 응시하며 휴대전화를 귀에 댄 채 입을 살짝 벌리고 멈춰 선 수리영의 상체.\n\nLOCATION (lock): A dog shelter with enclosed kennels for multiple dogs and practical areas for veterinary injections, grooming, treatment, and washing. 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 kennel-row look, metal-bar materials, daylight, and the young woman's clothing. Keep her in the same general position outside the cages while adding the phone at her ear. Exclude the earlier direct standoff with the aggressive dog and exclude the veterinarian if outside this frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has her mobile phone with her and is using it to call Minsuk, who does not answer.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 갇힌 개들을 응시하며 휴대전화를 귀에 댄 채 입을 살짝 벌리고 멈춰 선 수리영의 상체.\n\nLOCATION (lock): A dog shelter with enclosed kennels for multiple dogs and practical areas for veterinary injections, grooming, treatment, and washing. 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 kennel-row look, metal-bar materials, daylight, and the young woman's clothing. Keep her in the same general position outside the cages while adding the phone at her ear. Exclude the earlier direct standoff with the aggressive dog and exclude the veterinarian if outside this frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has her mobile phone with her and is using it to call Minsuk, who does not answer.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her upper body in an intimate eye-level profile medium close-up, holding her face, parted lips, and phone hand in crisp focus while her fixed gaze leads across the frame. Let the metal kennel bars and contained dogs recede behind her in shallow, layered depth, with soft directional daylight shaping her young face and the worn metal surfaces. Keep the composition quiet and suspended, emphasizing the unanswered-call stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 갇힌 개들을 응시하며 휴대전화를 귀에 댄 채 입을 살짝 벌리고 멈춰 선 수리영의 상체.\n\nLOCATION (lock): A dog shelter with enclosed kennels for multiple dogs and practical areas for veterinary injections, grooming, treatment, and washing. 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 kennel-row look, metal-bar materials, daylight, and the young woman's clothing. Keep her in the same general position outside the cages while adding the phone at her ear. Exclude the earlier direct standoff with the aggressive dog and exclude the veterinarian if outside this frame.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young has her mobile phone with her and is using it to call Minsuk, who does not answer.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera at a low kennel-side viewpoint, looking through prominent out-of-focus metal bars toward her upper body beyond them. Use a restrained wider lens feel and deeper spatial layering so the enclosure dominates the foreground while her halted posture and slightly open mouth remain clearly readable in the daylight. Compose her between the bars, making the physical separation from the dogs the central visual tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 8, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시사항에 따라 이전 컷의 공격적인 개를 배제하고, 갇힌 개들을 바라보며 통화하는 수리영의 상반신을 적절한 구도와 자세로 잘 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스의 카메라 구도를 그대로 복사했으며, 프롬프트에서 명시적으로 제외하라고 지시한 전경의 개를 그대로 포함하는 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/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/scene/recipe/S7sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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": "휴대전화의 화면이 사용자의 귀 쪽이 아닌 바깥쪽 카메라를 향하도록 잘못 뒤집혀 있습니다.", "fix_en": "Rotate the mobile phone in her hand so the screen faces her ear and the back of the phone faces outward." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Rotate the mobile phone in her hand so the screen faces her ear and the back of the phone faces outward.\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.", "ref_mode": "prev+엔티티", "lane_policy": "ab_select_bypass:prev" }, "S8sh1::variants": { "author_fp": "1c74907563ecc639", "author": { "variants": [ { "approach_ko": "허리 높이의 측면 트래킹으로 수리영은 선명하게 붙잡고 주변 네온과 포장마차는 흐르게 만든 역동적 접근.", "prompt_en": "Photograph the beat as a medium-wide lateral tracking shot from roughly waist height, keeping Suri-young and the bicycle crisply legible while the stalls, taxis, and neon-lit storefronts slide into restrained horizontal motion blur. Use a natural wide-angle perspective with layered foreground stall edges briefly framing her passage, and let the cool dusk ambience mix with warm food-stall light and colored neon across her face and trailing clothing." }, { "approach_ko": "포장마차 통로 끝의 낮은 정면 시점과 압축된 원근으로 수리영이 번화가를 가르며 다가오는 순간을 긴장감 있게 포착.", "prompt_en": "Place the camera low at the far end of the passage for a compressed, near-frontal view, with the stall fronts forming a tight visual corridor around Suri-young. Hold the frame steady and time the image at the most graphic point of the pedal stroke, keeping her hands, focused young face, bicycle geometry, and wind-pulled clothing sharply defined against softly layered neon signs and harbor-district traffic. Shape the dusk light as a cool base with directional neon reflections and warm stall spill creating depth rather than motion blur." } ] }, "reused": false }, "S8sh1": { "input_fingerprint": "6dfa44d4a3c77808", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 네온사인이 켜진 해질녘 번화가, 수리영이 자전거 페달을 밟는 중간 자세로 핸들을 쥔 채 포장마차 사이를 지나는 순간, 옷자락이 뒤로 흩날리고 있다.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young is still traveling with the same bicycle as she rides through the commercial district.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 네온사인이 켜진 해질녘 번화가, 수리영이 자전거 페달을 밟는 중간 자세로 핸들을 쥔 채 포장마차 사이를 지나는 순간, 옷자락이 뒤로 흩날리고 있다.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young is still traveling with the same bicycle as she rides through the commercial district.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the beat as a medium-wide lateral tracking shot from roughly waist height, keeping Suri-young and the bicycle crisply legible while the stalls, taxis, and neon-lit storefronts slide into restrained horizontal motion blur. Use a natural wide-angle perspective with layered foreground stall edges briefly framing her passage, and let the cool dusk ambience mix with warm food-stall light and colored neon across her face and trailing clothing.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 네온사인이 켜진 해질녘 번화가, 수리영이 자전거 페달을 밟는 중간 자세로 핸들을 쥔 채 포장마차 사이를 지나는 순간, 옷자락이 뒤로 흩날리고 있다.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young is still traveling with the same bicycle as she rides through the commercial district.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera low at the far end of the passage for a compressed, near-frontal view, with the stall fronts forming a tight visual corridor around Suri-young. Hold the frame steady and time the image at the most graphic point of the pedal stroke, keeping her hands, focused young face, bicycle geometry, and wind-pulled clothing sharply defined against softly layered neon signs and harbor-district traffic. Shape the dusk light as a cool base with directional neon reflections and warm stall spill creating depth rather than motion blur.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 네온사인이 켜진 해질녘 번화가, 수리영이 자전거 페달을 밟는 중간 자세로 핸들을 쥔 채 포장마차 사이를 지나는 순간, 옷자락이 뒤로 흩날리고 있다.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young is still traveling with the same bicycle as she rides through the commercial district.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the beat as a medium-wide lateral tracking shot from roughly waist height, keeping Suri-young and the bicycle crisply legible while the stalls, taxis, and neon-lit storefronts slide into restrained horizontal motion blur. Use a natural wide-angle perspective with layered foreground stall edges briefly framing her passage, and let the cool dusk ambience mix with warm food-stall light and colored neon across her face and trailing clothing.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 네온사인이 켜진 해질녘 번화가, 수리영이 자전거 페달을 밟는 중간 자세로 핸들을 쥔 채 포장마차 사이를 지나는 순간, 옷자락이 뒤로 흩날리고 있다.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young is still traveling with the same bicycle as she rides through the commercial district.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera low at the far end of the passage for a compressed, near-frontal view, with the stall fronts forming a tight visual corridor around Suri-young. Hold the frame steady and time the image at the most graphic point of the pedal stroke, keeping her hands, focused young face, bicycle geometry, and wind-pulled clothing sharply defined against softly layered neon signs and harbor-district traffic. Shape the dusk light as a cool base with directional neon reflections and warm stall spill creating depth rather than motion blur.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S8sh1.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S8sh1.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "D", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "로케이션 형태와 탑승 동작, 흩날리는 옷자락을 정확히 구현했으나 레퍼런스의 헬멧이 누락됨." }, { "label": "B", "score": 4, "verdict_ko": "자전거 앞바퀴 구조가 물리적으로 불가능하며, 포장마차를 임의로 추가해 로케이션을 훼손함." }, { "label": "C", "score": 3, "verdict_ko": "핸들을 쥔 왼팔이 보이지 않는 심각한 구조적 오류가 있으며, 외투 형태와 얼굴 조명이 왜곡됨." }, { "label": "D", "score": 5, "verdict_ko": "캐릭터의 착장과 헬멧은 일치하나, 마트가 사라지고 좁은 골목길로 로케이션이 전면 변형됨." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "D", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "로케이션 형태와 탑승 동작, 흩날리는 옷자락을 정확히 구현했으나 레퍼런스의 헬멧이 누락됨." }, { "label": "B", "score": 4, "verdict_ko": "자전거 앞바퀴 구조가 물리적으로 불가능하며, 포장마차를 임의로 추가해 로케이션을 훼손함." }, { "label": "C", "score": 3, "verdict_ko": "핸들을 쥔 왼팔이 보이지 않는 심각한 구조적 오류가 있으며, 외투 형태와 얼굴 조명이 왜곡됨." }, { "label": "D", "score": 5, "verdict_ko": "캐릭터의 착장과 헬멧은 일치하나, 마트가 사라지고 좁은 골목길로 로케이션이 전면 변형됨." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "포장마차 사이를 지나는 동선, 옷자락 흩날림, 헬멧 착용 등 핵심 지시를 가장 잘 충족했으나 크로스백 착용 방향이 반전됨." }, { "label": "B", "score": 6, "verdict_ko": "기준 장소와 바람에 흩날리는 옷자락 묘사는 우수하나, 포장마차 사이를 지나지 않으며 크로스백이 누락됨." }, { "label": "C", "score": 4, "verdict_ko": "필수 소품인 헬멧이 누락되었으며, 배경의 건축물이 기준 사진과 다르게 임의로 변형됨." }, { "label": "D", "score": 5, "verdict_ko": "동선과 장소 재현, 크로스백 묘사는 적절하나 기준 이미지에 명시된 헬멧이 누락됨." } ] }, "reverse_normalized": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "포장마차 사이를 지나는 동선, 옷자락 흩날림, 헬멧 착용 등 핵심 지시를 가장 잘 충족했으나 크로스백 착용 방향이 반전됨." }, { "label": "C", "score": 6, "verdict_ko": "기준 장소와 바람에 흩날리는 옷자락 묘사는 우수하나, 포장마차 사이를 지나지 않으며 크로스백이 누락됨." }, { "label": "B", "score": 4, "verdict_ko": "필수 소품인 헬멧이 누락되었으며, 배경의 건축물이 기준 사진과 다르게 임의로 변형됨." }, { "label": "A", "score": 5, "verdict_ko": "동선과 장소 재현, 크로스백 묘사는 적절하나 기준 이미지에 명시된 헬멧이 누락됨." } ] }, "combined": { "totals": { "A": 12, "B": 8, "C": 9, "D": 12 }, "ranking": [ "A", "D", "C", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 12, "B": 8, "C": 9, "D": 12 }, "selected": "A", "ranking": [ "A", "D", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "로케이션 형태와 탑승 동작, 흩날리는 옷자락을 정확히 구현했으나 레퍼런스의 헬멧이 누락됨." }, { "label": "B", "score": 4, "verdict_ko": "자전거 앞바퀴 구조가 물리적으로 불가능하며, 포장마차를 임의로 추가해 로케이션을 훼손함." }, { "label": "C", "score": 3, "verdict_ko": "핸들을 쥔 왼팔이 보이지 않는 심각한 구조적 오류가 있으며, 외투 형태와 얼굴 조명이 왜곡됨." }, { "label": "D", "score": 5, "verdict_ko": "캐릭터의 착장과 헬멧은 일치하나, 마트가 사라지고 좁은 골목길로 로케이션이 전면 변형됨." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "샷 텍스트에 명시되지 않은 수많은 행인들이 배경과 포장마차 주변에 추가되었습니다.", "fix_en": "Remove all pedestrians and other people from the entire image so that only Suri-young is visible." }, { "issue_ko": "프레임 안에 머리가 보임에도 불구하고 캐릭터 레퍼런스에 있는 검은색 자전거 헬멧을 착용하지 않았습니다.", "fix_en": "Add the black bicycle helmet to Suri-young's head, matching the character reference perfectly." }, { "issue_ko": "시각적 접근(Visual Approach)에서 요구한 배경(포장마차, 택시, 상점)의 수평 모션 블러 효과가 적용되지 않고 선명하게 정지되어 있습니다.", "fix_en": "Apply a restrained horizontal motion blur to the stalls, taxis, and storefronts in the background to convey a lateral tracking shot." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all pedestrians and other people from the entire image so that only Suri-young is visible.\n- Add the black bicycle helmet to Suri-young's head, matching the character reference perfectly.\n- Apply a restrained horizontal motion blur to the stalls, taxis, and storefronts in the background to convey a lateral tracking shot.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·A=변형0)", "plate_select": { "candidates": { "A": "L08B01", "B": "L08B02", "C": "L08B03", "D": "L08B04" }, "assigned": "L08B01", "choice": "Candidate A", "confident": true, "reason_ko": "지문에서 '네온사인이 켜진 해질녘 번화가'와 '포장마차 사이를 지나는 순간'이라고 명시하고 있으므로, 거리의 포장마차와 조명, 자전거가 주차된 마트 앞 풍경이 모두 포함된 A가 가장 적합합니다.", "kept": "L08B01" } }, "S8sh4::variants": { "author_fp": "ff5699beb2ecee6d", "author": { "variants": [ { "approach_ko": "눈높이의 밀착된 중근접 촬영으로 휴대전화 너머 수리영의 심각한 표정과 정지된 순간을 압축한다.", "prompt_en": "Photograph the moment in an intimate eye-level medium close-up, with the phone held naturally between Suri-young and the camera so its screen faces her and only its back and edge are visible. Keep her fixed gaze and restrained facial tension sharply legible while the harbor-side commercial lights fall into soft, layered neon blur behind her. Use a compressed, shallow-focus composition that makes her sudden stillness feel sealed off from the surrounding bustle." }, { "approach_ko": "넓고 약간 높은 시점의 환경 쇼트로 마트와 주차된 자전거 사이에 멈춘 수리영의 고립감을 강조한다.", "prompt_en": "Use a wide environmental composition from a slightly elevated viewpoint, holding Suri-young as a solitary stopped figure within the depth of the roadside setting. Arrange the parked bicycle and supermarket frontage as clear spatial anchors while taxis, stalls, and neon-lit harbor commerce form receding layers without overpowering her. Favor deeper focus and cool dusk ambience broken by directional neon reflections, emphasizing the contrast between the active surroundings and her grave, motionless attention to the phone." } ] }, "reused": false }, "S8sh4": { "input_fingerprint": "38136479febd9c53", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 길가에서 휴대전화 액정을 심각한 표정으로 들여다보며 멈춰 선 수리영.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young’s bicycle remains parked outside the mart while she uses her mobile phone to try Minsuk again.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 길가에서 휴대전화 액정을 심각한 표정으로 들여다보며 멈춰 선 수리영.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young’s bicycle remains parked outside the mart while she uses her mobile phone to try Minsuk again.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in an intimate eye-level medium close-up, with the phone held naturally between Suri-young and the camera so its screen faces her and only its back and edge are visible. Keep her fixed gaze and restrained facial tension sharply legible while the harbor-side commercial lights fall into soft, layered neon blur behind her. Use a compressed, shallow-focus composition that makes her sudden stillness feel sealed off from the surrounding bustle.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 길가에서 휴대전화 액정을 심각한 표정으로 들여다보며 멈춰 선 수리영.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young’s bicycle remains parked outside the mart while she uses her mobile phone to try Minsuk again.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wide environmental composition from a slightly elevated viewpoint, holding Suri-young as a solitary stopped figure within the depth of the roadside setting. Arrange the parked bicycle and supermarket frontage as clear spatial anchors while taxis, stalls, and neon-lit harbor commerce form receding layers without overpowering her. Favor deeper focus and cool dusk ambience broken by directional neon reflections, emphasizing the contrast between the active surroundings and her grave, motionless attention to the phone.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 길가에서 휴대전화 액정을 심각한 표정으로 들여다보며 멈춰 선 수리영.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young’s bicycle remains parked outside the mart while she uses her mobile phone to try Minsuk again.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in an intimate eye-level medium close-up, with the phone held naturally between Suri-young and the camera so its screen faces her and only its back and edge are visible. Keep her fixed gaze and restrained facial tension sharply legible while the harbor-side commercial lights fall into soft, layered neon blur behind her. Use a compressed, shallow-focus composition that makes her sudden stillness feel sealed off from the surrounding bustle.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 길가에서 휴대전화 액정을 심각한 표정으로 들여다보며 멈춰 선 수리영.\n\nLOCATION (lock): A busy commercial district beside a harbor, lined with neon signs, taxis, hawkers, and street-food stalls. A large supermarket with bicycle parking and stocked retail aisles anchors the area. 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): Suri-young’s bicycle remains parked outside the mart while she uses her mobile phone to try Minsuk again.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wide environmental composition from a slightly elevated viewpoint, holding Suri-young as a solitary stopped figure within the depth of the roadside setting. Arrange the parked bicycle and supermarket frontage as clear spatial anchors while taxis, stalls, and neon-lit harbor commerce form receding layers without overpowering her. Favor deeper focus and cool dusk ambience broken by directional neon reflections, emphasizing the contrast between the active surroundings and her grave, motionless attention to the phone.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B02.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S8sh4.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B02.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S8sh4.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "휴대전화 뒷면에 화면 텍스트가 나타나는 물리적 오류(하드 위반)가 있습니다." }, { "label": "B", "score": 6, "verdict_ko": "장소는 정확하나 지정된 의상(힙색)이 누락되었고 항구 배경이 보이지 않습니다." }, { "label": "C", "score": 5, "verdict_ko": "레퍼런스의 건축물 및 거리 구조가 임의로 변형되어 위치 고정 지시를 위반했습니다." }, { "label": "D", "score": 8, "verdict_ko": "장소, 항구 배경, 의상(힙색), 소품 방향을 모두 정확히 구현한 우수작입니다." } ] }, "forward_normalized": { "winner": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "휴대전화 뒷면에 화면 텍스트가 나타나는 물리적 오류(하드 위반)가 있습니다." }, { "label": "B", "score": 6, "verdict_ko": "장소는 정확하나 지정된 의상(힙색)이 누락되었고 항구 배경이 보이지 않습니다." }, { "label": "C", "score": 5, "verdict_ko": "레퍼런스의 건축물 및 거리 구조가 임의로 변형되어 위치 고정 지시를 위반했습니다." }, { "label": "D", "score": 8, "verdict_ko": "장소, 항구 배경, 의상(힙색), 소품 방향을 모두 정확히 구현한 우수작입니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "C", "B", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "장소, 복장, 자전거 배치 등 모든 지시사항을 가장 충실히 구현함." }, { "label": "B", "score": 4, "verdict_ko": "레퍼런스와 배경 건축물이 불일치하며 폰에 임의의 로고가 추가됨." }, { "label": "C", "score": 5, "verdict_ko": "배경은 원본과 같으나 지정된 초록색 가방이 누락됨." }, { "label": "D", "score": 3, "verdict_ko": "스마트폰 뒷면에 텍스트가 유출되는 결정적 오류가 있음." } ] }, "reverse_normalized": { "winner": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "장소, 복장, 자전거 배치 등 모든 지시사항을 가장 충실히 구현함." }, { "label": "C", "score": 4, "verdict_ko": "레퍼런스와 배경 건축물이 불일치하며 폰에 임의의 로고가 추가됨." }, { "label": "B", "score": 5, "verdict_ko": "배경은 원본과 같으나 지정된 초록색 가방이 누락됨." }, { "label": "A", "score": 3, "verdict_ko": "스마트폰 뒷면에 텍스트가 유출되는 결정적 오류가 있음." } ] }, "combined": { "totals": { "A": 6, "B": 11, "C": 9, "D": 15 }, "ranking": [ "D", "B", "C", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 11, "C": 9, "D": 15 }, "selected": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "휴대전화 뒷면에 화면 텍스트가 나타나는 물리적 오류(하드 위반)가 있습니다." }, { "label": "B", "score": 6, "verdict_ko": "장소는 정확하나 지정된 의상(힙색)이 누락되었고 항구 배경이 보이지 않습니다." }, { "label": "C", "score": 5, "verdict_ko": "레퍼런스의 건축물 및 거리 구조가 임의로 변형되어 위치 고정 지시를 위반했습니다." }, { "label": "D", "score": 8, "verdict_ko": "장소, 항구 배경, 의상(힙색), 소품 방향을 모두 정확히 구현한 우수작입니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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": "샷 텍스트에 수리영만 등장해야 함에도 불구하고 배경 거리와 노점에 허구의 인물들이 다수 추가되었습니다.", "fix_en": "Remove all background people, pedestrians, and vendors from the street and stalls, leaving only Suri-young visible." }, { "issue_ko": "'텍스트 없음' 규칙을 위반하여 오른쪽 거리의 네온사인과 노점 간판들에 문자가 생성되어 있습니다.", "fix_en": "Remove all text and lettering from the neon signs and stall banners on the right side of the street." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all background people, pedestrians, and vendors from the street and stalls, leaving only Suri-young visible.\n- Remove all text and lettering from the neon signs and stall banners on the right side of the street.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+엔티티 (4택1: 무콘티 승·D=변형1)", "plate_select": { "candidates": { "A": "L08B01", "B": "L08B02", "C": "L08B03", "D": "L08B04" }, "assigned": "L08B02", "choice": "Candidate B", "confident": true, "reason_ko": "지문에서 '마트 밖 길가에서'라고만 명시되어 있어, 마트 외부 길가를 보여주는 후보 A와 B 중 어느 한 곳으로 특정할 만한 명확한 단서(포장마차, 자전거 등)가 없으므로 현재 할당된 후보를 유지합니다.", "kept": "L08B02" } }, "S9sh1::variants": { "author_fp": "b72794dc2f6648d1", "author": { "variants": [ { "approach_ko": "자전거 높이의 넓은 측면 숏으로 밤바다의 여백과 수리영의 다급한 페달 동작을 함께 강조한다.", "prompt_en": "Photograph the moment in a broad, clean side profile from around bicycle height, holding Suri-young and the bicycle clearly against the dark sea. Use strong lateral composition and generous shoreline negative space, with her bent knee, forceful downward pedal stroke, rigid expression, and backward-streaming hair and clothing all legible in one kinetic figure. Keep the low-key night image crisp on her while allowing only restrained directional motion in the background." }, { "approach_ko": "수리영과 나란히 달리는 밀착 측면 카메라로 굳은 얼굴과 온몸의 추진력을 압축해 포착한다.", "prompt_en": "Track tightly beside Suri-young at near eye level in an intimate side-on framing, compressing the dark sea behind her and concentrating attention on her tense face and driving posture. Frame closely enough to make the backward sweep of her hair and clothing feel urgent while still preserving the bent knee and pedal at the decisive midpoint of the push. Use shallow depth and soft, low-key night illumination, with her face and pedaling motion held sharply against the subdued background." } ] }, "reused": false }, "S9sh1": { "input_fingerprint": "2a64b88803ef89a7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다를 배경으로, 수리영이 굳은 표정으로 한쪽 무릎을 굽혀 자전거 페달을 힘껏 밟아 내리는 중간 동작, 옷자락과 머리카락이 뒤로 흩날리는 측면.\n\nLOCATION (lock): A coastal road tracing the nighttime shoreline, used as a bicycle route back toward the residential district. 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): Suri-young is again riding her bicycle, now hurrying home along the coastal road.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다를 배경으로, 수리영이 굳은 표정으로 한쪽 무릎을 굽혀 자전거 페달을 힘껏 밟아 내리는 중간 동작, 옷자락과 머리카락이 뒤로 흩날리는 측면.\n\nLOCATION (lock): A coastal road tracing the nighttime shoreline, used as a bicycle route back toward the residential district. 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): Suri-young is again riding her bicycle, now hurrying home along the coastal road.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in a broad, clean side profile from around bicycle height, holding Suri-young and the bicycle clearly against the dark sea. Use strong lateral composition and generous shoreline negative space, with her bent knee, forceful downward pedal stroke, rigid expression, and backward-streaming hair and clothing all legible in one kinetic figure. Keep the low-key night image crisp on her while allowing only restrained directional motion in the background.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다를 배경으로, 수리영이 굳은 표정으로 한쪽 무릎을 굽혀 자전거 페달을 힘껏 밟아 내리는 중간 동작, 옷자락과 머리카락이 뒤로 흩날리는 측면.\n\nLOCATION (lock): A coastal road tracing the nighttime shoreline, used as a bicycle route back toward the residential district. 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): Suri-young is again riding her bicycle, now hurrying home along the coastal road.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTrack tightly beside Suri-young at near eye level in an intimate side-on framing, compressing the dark sea behind her and concentrating attention on her tense face and driving posture. Frame closely enough to make the backward sweep of her hair and clothing feel urgent while still preserving the bent knee and pedal at the decisive midpoint of the push. Use shallow depth and soft, low-key night illumination, with her face and pedaling motion held sharply against the subdued background.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 측면 구도와 스토리보드 배치를 정확히 따랐으며, 헬멧을 포함한 인물 레퍼런스의 착장을 충실히 반영했습니다." }, { "label": "B", "score": 3, "verdict_ko": "요구된 측면(프로필) 앵글이 아닌 사선 구도로 촬영되었고, 필수 의상인 헬멧이 누락되었으며 페달 주변 발의 형태가 물리적으로 어색합니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S9sh1.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물의 바지가 레퍼런스의 일자 청바지가 아닌, 발목에 밴드가 있는 파란색 트레이닝 바지로 생성되었습니다.", "fix_en": "Change the pants to straight-leg blue denim jeans exactly matching the character reference." }, { "issue_ko": "신발이 레퍼런스 이미지의 두꺼운 등산화 스타일이 아닌, 굽이 낮은 단화로 생성되었습니다.", "fix_en": "Change the shoes to the chunky black and grey hiking sneakers shown in the character reference." }, { "issue_ko": "페달을 밟고 있는 왼쪽 발 아래에 다리와 연결되지 않은 여분의 신발과 페달 형태가 하나 더 생성되었습니다.", "fix_en": "Remove the extra unattached shoe floating below the main pedal." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the pants to straight-leg blue denim jeans exactly matching the character reference.\n- Change the shoes to the chunky black and grey hiking sneakers shown in the character reference.\n- Remove the extra unattached shoe floating below the main pedal.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S10sh3::variants": { "author_fp": "b4c55786568be156", "author": { "variants": [ { "approach_ko": "철문 틈과 멈춘 손가락을 정면 초근접으로 압축해 잠금 해제의 긴장과 촉감을 강조한다.", "prompt_en": "Photograph the action in an extremely tight, nearly straight-on close-up, cropping decisively around the hand, the compressed flyer, and the narrow seam of the iron door. Keep the door surface almost parallel to the image plane so the rigid geometry of the sealed entrance contrasts with the fingers held under pressure; use shallow focus centered on the exact point of contact, with restrained nighttime falloff across the surrounding metal." }, { "approach_ko": "철문 표면을 따라 비스듬히 파고드는 측면 근접 구도로 손과 틈새의 깊이, 눌린 전단지의 압력을 드러낸다.", "prompt_en": "Use a close, sharply oblique side angle that looks along the iron door surface toward the hand and its point of entry into the seam. Let the near metal edge dominate the foreground and fall out of focus, while the fingers and pressed flyer sit in a narrow plane of crisp detail deeper in the frame; low raking night light should reveal the door texture, the tension in the fingertips, and the compressed layers at the gap." } ] }, "reused": false }, "S10sh3": { "input_fingerprint": "dbfcaf253f062bb5", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 굳게 닫힌 옥탑방 철문 틈새로 손가락을 밀어 넣어 전단지를 꾹 누른 채 멈춘 한국인 남성의 손 클로즈업.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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 advertising flyer has been removed from the door gap and is being pressed into the gap to release the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 굳게 닫힌 옥탑방 철문 틈새로 손가락을 밀어 넣어 전단지를 꾹 누른 채 멈춘 한국인 남성의 손 클로즈업.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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 advertising flyer has been removed from the door gap and is being pressed into the gap to release the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the action in an extremely tight, nearly straight-on close-up, cropping decisively around the hand, the compressed flyer, and the narrow seam of the iron door. Keep the door surface almost parallel to the image plane so the rigid geometry of the sealed entrance contrasts with the fingers held under pressure; use shallow focus centered on the exact point of contact, with restrained nighttime falloff across the surrounding metal.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 굳게 닫힌 옥탑방 철문 틈새로 손가락을 밀어 넣어 전단지를 꾹 누른 채 멈춘 한국인 남성의 손 클로즈업.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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 advertising flyer has been removed from the door gap and is being pressed into the gap to release the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close, sharply oblique side angle that looks along the iron door surface toward the hand and its point of entry into the seam. Let the near metal edge dominate the foreground and fall out of focus, while the fingers and pressed flyer sit in a narrow plane of crisp detail deeper in the frame; low raking night light should reveal the door texture, the tension in the fingertips, and the compressed layers at the gap.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 굳게 닫힌 옥탑방 철문 틈새로 손가락을 밀어 넣어 전단지를 꾹 누른 채 멈춘 한국인 남성의 손 클로즈업.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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 advertising flyer has been removed from the door gap and is being pressed into the gap to release the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the action in an extremely tight, nearly straight-on close-up, cropping decisively around the hand, the compressed flyer, and the narrow seam of the iron door. Keep the door surface almost parallel to the image plane so the rigid geometry of the sealed entrance contrasts with the fingers held under pressure; use shallow focus centered on the exact point of contact, with restrained nighttime falloff across the surrounding metal.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 굳게 닫힌 옥탑방 철문 틈새로 손가락을 밀어 넣어 전단지를 꾹 누른 채 멈춘 한국인 남성의 손 클로즈업.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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 advertising flyer has been removed from the door gap and is being pressed into the gap to release the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close, sharply oblique side angle that looks along the iron door surface toward the hand and its point of entry into the seam. Let the near metal edge dominate the foreground and fall out of focus, while the fingers and pressed flyer sit in a narrow plane of crisp detail deeper in the frame; low raking night light should reveal the door texture, the tension in the fingertips, and the compressed layers at the gap.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S10sh3.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S10sh3.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "D", "B", "C" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "'손가락을 철문 틈새로 밀어 넣어 전단지를 누르는' 핵심 동작과 의상을 가장 정확하게 묘사하여 높은 완성도를 보입니다." }, { "label": "B", "score": 5, "verdict_ko": "전단지를 누르지 않고 문 주변에 손을 두고 있어 '틈새로 손가락을 밀어 넣는' 주요 동작 지시를 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "철문 틈새가 아닌 벽면에 전단지를 대고 있으며, 금지된 텍스트가 뚜렷하게 생성되어 크게 감점되었습니다." }, { "label": "D", "score": 6, "verdict_ko": "구조물 참조 사진의 녹색 문을 반영했으나, 손가락이 틈새로 들어가는 묘사가 부족하고 전단지에 텍스트가 포함되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "D", "B", "C" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "'손가락을 철문 틈새로 밀어 넣어 전단지를 누르는' 핵심 동작과 의상을 가장 정확하게 묘사하여 높은 완성도를 보입니다." }, { "label": "B", "score": 5, "verdict_ko": "전단지를 누르지 않고 문 주변에 손을 두고 있어 '틈새로 손가락을 밀어 넣는' 주요 동작 지시를 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "철문 틈새가 아닌 벽면에 전단지를 대고 있으며, 금지된 텍스트가 뚜렷하게 생성되어 크게 감점되었습니다." }, { "label": "D", "score": 6, "verdict_ko": "구조물 참조 사진의 녹색 문을 반영했으나, 손가락이 틈새로 들어가는 묘사가 부족하고 전단지에 텍스트가 포함되었습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "C", "B", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "철문 틈새로 전단지를 누르는 동작과 지정된 장소의 디테일이 가장 정확하게 묘사됨." }, { "label": "C", "score": 5, "verdict_ko": "장소와 의상은 일치하나 전단지를 틈새로 누르는 핵심 손가락 동작이 구현되지 않음." }, { "label": "B", "score": 4, "verdict_ko": "틈새가 아닌 평면에 손을 대고 있으며, 전단지에 금지된 텍스트가 노출됨." }, { "label": "A", "score": 3, "verdict_ko": "레퍼런스의 회색 문이 아닌 임의의 녹색 문이 등장하여 장소 설정에 오류가 있음." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "철문 틈새로 전단지를 누르는 동작과 지정된 장소의 디테일이 가장 정확하게 묘사됨." }, { "label": "B", "score": 5, "verdict_ko": "장소와 의상은 일치하나 전단지를 틈새로 누르는 핵심 손가락 동작이 구현되지 않음." }, { "label": "C", "score": 4, "verdict_ko": "틈새가 아닌 평면에 손을 대고 있으며, 전단지에 금지된 텍스트가 노출됨." }, { "label": "D", "score": 3, "verdict_ko": "레퍼런스의 회색 문이 아닌 임의의 녹색 문이 등장하여 장소 설정에 오류가 있음." } ] }, "combined": { "totals": { "A": 16, "B": 10, "C": 7, "D": 9 }, "ranking": [ "A", "B", "D", "C" ], "agreement": true, "policy": 1 } }, "totals": { "A": 16, "B": 10, "C": 7, "D": 9 }, "selected": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "'손가락을 철문 틈새로 밀어 넣어 전단지를 누르는' 핵심 동작과 의상을 가장 정확하게 묘사하여 높은 완성도를 보입니다." }, { "label": "B", "score": 5, "verdict_ko": "전단지를 누르지 않고 문 주변에 손을 두고 있어 '틈새로 손가락을 밀어 넣는' 주요 동작 지시를 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "철문 틈새가 아닌 벽면에 전단지를 대고 있으며, 금지된 텍스트가 뚜렷하게 생성되어 크게 감점되었습니다." }, { "label": "D", "score": 6, "verdict_ko": "구조물 참조 사진의 녹색 문을 반영했으나, 손가락이 틈새로 들어가는 묘사가 부족하고 전단지에 텍스트가 포함되었습니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "검지손가락이 철문 틈새로 들어가 전단지를 누르지 않고 문 앞 표면에 얹혀 있습니다.", "fix_en": "Reposition the index finger so it is pushed directly into the door gap, actively pressing against the flyer." }, { "issue_ko": "철문의 색상이 'STRUCTURE LOOK' 레퍼런스에 지정된 초록색이 아닌 회색으로 렌더링되었습니다.", "fix_en": "Change the color of the iron door to dark green to match the rooftop structure in the STRUCTURE LOOK photograph." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Reposition the index finger so it is pushed directly into the door gap, actively pressing against the flyer.\n- Change the color of the iron door to dark green to match the rooftop structure in the STRUCTURE LOOK photograph.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+seed+엔티티 (복잡구조물 4택1: 콘티 승·A=변형0)", "lane_policy": "ab_select_ready", "plate_select": { "candidates": { "A": "L03B01", "B": "L03B02", "C": "L03B03" }, "assigned": "L03B03", "choice": "Candidate C", "confident": true, "reason_ko": "지문에서 '옥탑방 철문'을 명시하고 있으며, 제공된 후보들 모두 철문이 포함된 옥탑방 전체를 보여줍니다. 후보 간 서브공간의 차이가 없으므로 현재 할당된 후보 C를 유지합니다.", "kept": "L03B03" } }, "S10sh5::variants": { "author_fp": "387330d32167a639", "author": { "variants": [ { "approach_ko": "옥상 건너편의 관찰자 시점으로 창문과 주변 구조를 함께 담아, 창 안의 공격을 불길한 그림자극처럼 포착한다.", "prompt_en": "Photograph from a distant, level rooftop vantage with a compressed telephoto feel, making the illuminated rooftop-room window the visual anchor while retaining the surrounding rooftop structures as dark spatial layers. Inside the window, preserve recognizable human anatomy, clothing, and partial facial detail as the intruder surges into Minsuk’s shoulder mid-bite; use the interior backlight and depth alignment to cast his presence as an unnaturally large but physically plausible shadow across her. Hold the composition still and observational, with the violent overlap concentrated inside the bright rectangular window." }, { "approach_ko": "창문 가까이에서 비스듬히 올려다보는 밀착 구도로, 민숙의 어깨와 침입자의 덮치는 동작을 왜곡된 크기 차이로 강조한다.", "prompt_en": "Move close to the window at a low, steeply oblique angle with a broad, immersive lens feel, cropping tightly around the intersecting bodies and the lit glass. Place Minsuk’s shoulder and marked wrist near the foreground edge while the intruder lunges from deeper in the room, his proximity to the window making his fully formed human figure and cast shadow loom disproportionately over her without becoming supernatural. Let the interior light rake across their profiles, hands, clothing, and the point of impact, freezing the attack at its most kinetic instant with shallow, urgent depth." } ] }, "reused": false }, "S10sh5": { "input_fingerprint": "7685347234100272", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 불이 켜진 옥탑방 창문에 비친 실루엣, 거대한 그림자가 강민숙의 어깨를 물어뜯기 위해 덮치는 mid-action 순간.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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): Minsuk still bears the longstanding circular wrist mark as the intruder attacks and bites into her shoulder.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양); 강민숙 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 불이 켜진 옥탑방 창문에 비친 실루엣, 거대한 그림자가 강민숙의 어깨를 물어뜯기 위해 덮치는 mid-action 순간.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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): Minsuk still bears the longstanding circular wrist mark as the intruder attacks and bites into her shoulder.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양); 강민숙 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a distant, level rooftop vantage with a compressed telephoto feel, making the illuminated rooftop-room window the visual anchor while retaining the surrounding rooftop structures as dark spatial layers. Inside the window, preserve recognizable human anatomy, clothing, and partial facial detail as the intruder surges into Minsuk’s shoulder mid-bite; use the interior backlight and depth alignment to cast his presence as an unnaturally large but physically plausible shadow across her. Hold the composition still and observational, with the violent overlap concentrated inside the bright rectangular window.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 불이 켜진 옥탑방 창문에 비친 실루엣, 거대한 그림자가 강민숙의 어깨를 물어뜯기 위해 덮치는 mid-action 순간.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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): Minsuk still bears the longstanding circular wrist mark as the intruder attacks and bites into her shoulder.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양); 강민숙 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nMove close to the window at a low, steeply oblique angle with a broad, immersive lens feel, cropping tightly around the intersecting bodies and the lit glass. Place Minsuk’s shoulder and marked wrist near the foreground edge while the intruder lunges from deeper in the room, his proximity to the window making his fully formed human figure and cast shadow loom disproportionately over her without becoming supernatural. Let the interior light rake across their profiles, hands, clothing, and the point of impact, freezing the attack at its most kinetic instant with shallow, urgent depth.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 불이 켜진 옥탑방 창문에 비친 실루엣, 거대한 그림자가 강민숙의 어깨를 물어뜯기 위해 덮치는 mid-action 순간.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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): Minsuk still bears the longstanding circular wrist mark as the intruder attacks and bites into her shoulder.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양); 강민숙 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a distant, level rooftop vantage with a compressed telephoto feel, making the illuminated rooftop-room window the visual anchor while retaining the surrounding rooftop structures as dark spatial layers. Inside the window, preserve recognizable human anatomy, clothing, and partial facial detail as the intruder surges into Minsuk’s shoulder mid-bite; use the interior backlight and depth alignment to cast his presence as an unnaturally large but physically plausible shadow across her. Hold the composition still and observational, with the violent overlap concentrated inside the bright rectangular window.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 불이 켜진 옥탑방 창문에 비친 실루엣, 거대한 그림자가 강민숙의 어깨를 물어뜯기 위해 덮치는 mid-action 순간.\n\nLOCATION (lock): The rooftop of an old villa, fitted with a water tank, clotheslines, a low wooden platform holding dried herbs, and an iron entrance door to the rooftop dwelling. 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): Minsuk still bears the longstanding circular wrist mark as the intruder attacks and bites into her shoulder.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 전단지 남성 (한국인 성인 남성, 가려진 얼굴, 식별되지 않는 머리 모양); 강민숙 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nMove close to the window at a low, steeply oblique angle with a broad, immersive lens feel, cropping tightly around the intersecting bodies and the lit glass. Place Minsuk’s shoulder and marked wrist near the foreground edge while the intruder lunges from deeper in the room, his proximity to the window making his fully formed human figure and cast shadow loom disproportionately over her without becoming supernatural. Let the interior light rake across their profiles, hands, clothing, and the point of impact, freezing the attack at its most kinetic instant with shallow, urgent depth.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S10sh5.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S10sh5.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "창문 프레임이 건물 벽이 아닌 옥상 허공에 떠 있는 콜라주 형태의 치명적인 물리적 구조 오류가 발생했습니다." }, { "label": "B", "score": 5, "verdict_ko": "건물 구조와 표식은 잘 구현했으나, 습격자가 그림자나 실루엣이 아닌 밝게 노출된 상태로 묘사되어 프롬프트의 지시를 어겼습니다." }, { "label": "C", "score": 2, "verdict_ko": "A와 동일하게 창문이 허공에 분리되어 떠 있는 불가능한 배경 배치를 보여주어 심각한 규정 위반입니다." }, { "label": "D", "score": 8, "verdict_ko": "실제 벽면에 위치한 창문 구조를 정확히 구현했으며, 습격자를 짙은 그림자 형태로 연출하여 프롬프트의 묘사와 손목 표식을 모두 훌륭하게 살렸습니다." } ] }, "forward_normalized": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "창문 프레임이 건물 벽이 아닌 옥상 허공에 떠 있는 콜라주 형태의 치명적인 물리적 구조 오류가 발생했습니다." }, { "label": "B", "score": 5, "verdict_ko": "건물 구조와 표식은 잘 구현했으나, 습격자가 그림자나 실루엣이 아닌 밝게 노출된 상태로 묘사되어 프롬프트의 지시를 어겼습니다." }, { "label": "C", "score": 2, "verdict_ko": "A와 동일하게 창문이 허공에 분리되어 떠 있는 불가능한 배경 배치를 보여주어 심각한 규정 위반입니다." }, { "label": "D", "score": 8, "verdict_ko": "실제 벽면에 위치한 창문 구조를 정확히 구현했으며, 습격자를 짙은 그림자 형태로 연출하여 프롬프트의 묘사와 손목 표식을 모두 훌륭하게 살렸습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "물리적으로 자연스러운 공간 속에서 어깨를 무는 액션과 손목의 원형 표식을 가장 정확하게 구현했습니다." }, { "label": "B", "score": 1, "verdict_ko": "옥상 배경 위에 창문 프레임이 공중에 떠 있는 단순 합성(콜라주) 형태이므로 치명적인 하드 위반입니다." }, { "label": "C", "score": 4, "verdict_ko": "공간 연출은 자연스러우나, 어깨를 무는 핵심 액션 대신 목을 조르고 있어 지시를 위반했습니다." }, { "label": "D", "score": 2, "verdict_ko": "실루엣 연출을 시도했으나, B와 마찬가지로 배경 위에 창문이 떠 있는 불가능한 합성 구조라 실격입니다." } ] }, "reverse_normalized": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "물리적으로 자연스러운 공간 속에서 어깨를 무는 액션과 손목의 원형 표식을 가장 정확하게 구현했습니다." }, { "label": "C", "score": 1, "verdict_ko": "옥상 배경 위에 창문 프레임이 공중에 떠 있는 단순 합성(콜라주) 형태이므로 치명적인 하드 위반입니다." }, { "label": "B", "score": 4, "verdict_ko": "공간 연출은 자연스러우나, 어깨를 무는 핵심 액션 대신 목을 조르고 있어 지시를 위반했습니다." }, { "label": "A", "score": 2, "verdict_ko": "실루엣 연출을 시도했으나, B와 마찬가지로 배경 위에 창문이 떠 있는 불가능한 합성 구조라 실격입니다." } ] }, "combined": { "totals": { "A": 5, "B": 9, "C": 3, "D": 15 }, "ranking": [ "D", "B", "A", "C" ], "agreement": true, "policy": 1 } }, "totals": { "A": 5, "B": 9, "C": 3, "D": 15 }, "selected": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "창문 프레임이 건물 벽이 아닌 옥상 허공에 떠 있는 콜라주 형태의 치명적인 물리적 구조 오류가 발생했습니다." }, { "label": "B", "score": 5, "verdict_ko": "건물 구조와 표식은 잘 구현했으나, 습격자가 그림자나 실루엣이 아닌 밝게 노출된 상태로 묘사되어 프롬프트의 지시를 어겼습니다." }, { "label": "C", "score": 2, "verdict_ko": "A와 동일하게 창문이 허공에 분리되어 떠 있는 불가능한 배경 배치를 보여주어 심각한 규정 위반입니다." }, { "label": "D", "score": 8, "verdict_ko": "실제 벽면에 위치한 창문 구조를 정확히 구현했으며, 습격자를 짙은 그림자 형태로 연출하여 프롬프트의 묘사와 손목 표식을 모두 훌륭하게 살렸습니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 전단지 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "남성 침입자의 얼굴과 머리 모양이 선명하게 보여 '가려진 얼굴, 식별되지 않는 머리 모양' 지침을 위반했습니다.", "fix_en": "Cast the male intruder's head in deep shadow to completely obscure his facial features and hairstyle." }, { "issue_ko": "창문이 너무 크고 세로로 길게 묘사되어, 레퍼런스 사진의 구조물에 있는 작고 높은 위치의 직사각형 창문 형태와 비율을 위반했습니다.", "fix_en": "Resize the window to match the small, rectangular proportions and higher wall placement seen on the structure in the reference photographs." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Cast the male intruder's head in deep shadow to completely obscure his facial features and hairstyle.\n- Resize the window to match the small, rectangular proportions and higher wall placement seen on the structure in the reference photographs.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+seed+엔티티 (복잡구조물 4택1: 무콘티 승·D=변형1)", "lane_policy": "ab_select_ready", "plate_select": { "candidates": { "A": "L03B01", "B": "L03B02", "C": "L03B03" }, "assigned": "L03B03", "choice": "Candidate C", "confident": true, "reason_ko": "샷 텍스트에 '불이 켜진 옥탑방 창문'이 명시되어 있으므로, 해당 창문이 강조된 서브공간이 필요합니다. 후보 C가 기존 할당본이며 조건을 만족합니다.", "kept": "L03B03" } }, "S11sh3::variants": { "author_fp": "e88c50995fa3983a", "author": { "variants": [ { "approach_ko": "수리영의 어깨를 크게 전경에 두고 열린 철문에만 시선을 압축하는 얕은 초점의 긴장감 있는 오버숄더 숏.", "prompt_en": "Photograph a tight over-the-shoulder view at her shoulder height, with her dark-haired head and shoulder occupying a substantial soft foreground edge while the iron entrance door holds crisp focus in the darkness. Use a compressed lens feel and restrained negative space around the doorway, letting the existing night illumination fall away quickly so the half-open gap becomes the dominant point of tension." }, { "approach_ko": "수리영의 어깨에서 계단과 옥탑방 문까지 이어지는 공간을 깊게 보여 주는 넓고 정적인 오버숄더 구도.", "prompt_en": "Use a wider, static over-the-shoulder composition that preserves the spatial path from her near shoulder through the exterior stairway to the rooftop entrance. Keep the layered architecture legible in deeper focus, placing her only as a narrow foreground anchor while the half-open door sits farther away within the surrounding darkness; favor natural perspective and the location’s existing night-light direction for a watchful, ominous distance." } ] }, "reused": false }, "S11sh3": { "input_fingerprint": "89345d9b542bc32f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 보이는, 어둠 속에서 반쯤 열린 채 멈춰 있는 옥탑방 현관 철문.\n\nLOCATION (lock): A narrow residential alley leading into the rear courtyard of a multi-unit villa. An exterior stairway climbs to a rooftop dwelling with a half-open iron door. 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 rooftop room’s iron entrance door remains half open after the intruder entered.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 보이는, 어둠 속에서 반쯤 열린 채 멈춰 있는 옥탑방 현관 철문.\n\nLOCATION (lock): A narrow residential alley leading into the rear courtyard of a multi-unit villa. An exterior stairway climbs to a rooftop dwelling with a half-open iron door. 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 rooftop room’s iron entrance door remains half open after the intruder entered.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph a tight over-the-shoulder view at her shoulder height, with her dark-haired head and shoulder occupying a substantial soft foreground edge while the iron entrance door holds crisp focus in the darkness. Use a compressed lens feel and restrained negative space around the doorway, letting the existing night illumination fall away quickly so the half-open gap becomes the dominant point of tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 보이는, 어둠 속에서 반쯤 열린 채 멈춰 있는 옥탑방 현관 철문.\n\nLOCATION (lock): A narrow residential alley leading into the rear courtyard of a multi-unit villa. An exterior stairway climbs to a rooftop dwelling with a half-open iron door. 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 rooftop room’s iron entrance door remains half open after the intruder entered.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider, static over-the-shoulder composition that preserves the spatial path from her near shoulder through the exterior stairway to the rooftop entrance. Keep the layered architecture legible in deeper focus, placing her only as a narrow foreground anchor while the half-open door sits farther away within the surrounding darkness; favor natural perspective and the location’s existing night-light direction for a watchful, ominous distance.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 보이는, 어둠 속에서 반쯤 열린 채 멈춰 있는 옥탑방 현관 철문.\n\nLOCATION (lock): A narrow residential alley leading into the rear courtyard of a multi-unit villa. An exterior stairway climbs to a rooftop dwelling with a half-open iron door. 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 rooftop room’s iron entrance door remains half open after the intruder entered.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph a tight over-the-shoulder view at her shoulder height, with her dark-haired head and shoulder occupying a substantial soft foreground edge while the iron entrance door holds crisp focus in the darkness. Use a compressed lens feel and restrained negative space around the doorway, letting the existing night illumination fall away quickly so the half-open gap becomes the dominant point of tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 보이는, 어둠 속에서 반쯤 열린 채 멈춰 있는 옥탑방 현관 철문.\n\nLOCATION (lock): A narrow residential alley leading into the rear courtyard of a multi-unit villa. An exterior stairway climbs to a rooftop dwelling with a half-open iron door. 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 rooftop room’s iron entrance door remains half open after the intruder entered.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider, static over-the-shoulder composition that preserves the spatial path from her near shoulder through the exterior stairway to the rooftop entrance. Keep the layered architecture legible in deeper focus, placing her only as a narrow foreground anchor while the half-open door sits farther away within the surrounding darkness; favor natural perspective and the location’s existing night-light direction for a watchful, ominous distance.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L10B03.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S11sh3.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L10B03.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S11sh3.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L10B03.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L10B03.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "C", "ranking": [ "C", "A", "B", "D" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "요구된 오버더숄더 앵글과 반쯤 열린 철문, 야간 배경을 잘 구현했으며 캐릭터 레퍼런스의 파란색 재킷 의상까지 충실히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 구도와 공간적 배경, 시간대는 잘 표현되었으나, 인물의 의상 색상이 레퍼런스와 일치하지 않습니다." }, { "label": "B", "score": 4, "verdict_ko": "구도와 배경은 적절하지만 인물이 프레임 바깥으로 너무 밀려나 있고 어두워서 의상과 형태를 식별하기 어렵습니다." }, { "label": "D", "score": 3, "verdict_ko": "인물이 프레임에 너무 많이 들어와 구도감이 떨어지며, 의상이 검은색 재킷으로 렌더링되어 레퍼런스와 크게 어긋납니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "A", "B", "D" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "요구된 오버더숄더 앵글과 반쯤 열린 철문, 야간 배경을 잘 구현했으며 캐릭터 레퍼런스의 파란색 재킷 의상까지 충실히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 구도와 공간적 배경, 시간대는 잘 표현되었으나, 인물의 의상 색상이 레퍼런스와 일치하지 않습니다." }, { "label": "B", "score": 4, "verdict_ko": "구도와 배경은 적절하지만 인물이 프레임 바깥으로 너무 밀려나 있고 어두워서 의상과 형태를 식별하기 어렵습니다." }, { "label": "D", "score": 3, "verdict_ko": "인물이 프레임에 너무 많이 들어와 구도감이 떨어지며, 의상이 검은색 재킷으로 렌더링되어 레퍼런스와 크게 어긋납니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A", "D", "C" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "오버더숄더 구도와 배경은 훌륭하나, 캐릭터의 의상(검은 재킷)과 긴 머리가 레퍼런스와 불일치합니다." }, { "label": "B", "score": 7, "verdict_ko": "요청된 오버더숄더 앵글과 반쯤 열린 철문을 정확히 연출했으며, 캐릭터의 파란색 재킷까지 잘 반영된 가장 우수한 결과물입니다." }, { "label": "C", "score": 4, "verdict_ko": "배경은 지정된 장소와 일치하지만, 전경의 인물이 단순한 어두운 실루엣으로만 묘사되어 캐릭터 식별이 어렵습니다." }, { "label": "D", "score": 4, "verdict_ko": "구도와 옥탑방 문의 상태는 맞으나, 인물의 의상이 레퍼런스(파란색 재킷)와 다르고 헤어스타일도 어긋납니다." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "D", "A", "B" ], "verdicts": [ { "label": "D", "score": 5, "verdict_ko": "오버더숄더 구도와 배경은 훌륭하나, 캐릭터의 의상(검은 재킷)과 긴 머리가 레퍼런스와 불일치합니다." }, { "label": "C", "score": 7, "verdict_ko": "요청된 오버더숄더 앵글과 반쯤 열린 철문을 정확히 연출했으며, 캐릭터의 파란색 재킷까지 잘 반영된 가장 우수한 결과물입니다." }, { "label": "B", "score": 4, "verdict_ko": "배경은 지정된 장소와 일치하지만, 전경의 인물이 단순한 어두운 실루엣으로만 묘사되어 캐릭터 식별이 어렵습니다." }, { "label": "A", "score": 4, "verdict_ko": "구도와 옥탑방 문의 상태는 맞으나, 인물의 의상이 레퍼런스(파란색 재킷)와 다르고 헤어스타일도 어긋납니다." } ] }, "combined": { "totals": { "A": 9, "B": 8, "C": 14, "D": 8 }, "ranking": [ "C", "A", "B", "D" ], "agreement": true, "policy": 1 } }, "totals": { "A": 9, "B": 8, "C": 14, "D": 8 }, "selected": "C", "ranking": [ "C", "A", "B", "D" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "요구된 오버더숄더 앵글과 반쯤 열린 철문, 야간 배경을 잘 구현했으며 캐릭터 레퍼런스의 파란색 재킷 의상까지 충실히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 구도와 공간적 배경, 시간대는 잘 표현되었으나, 인물의 의상 색상이 레퍼런스와 일치하지 않습니다." }, { "label": "B", "score": 4, "verdict_ko": "구도와 배경은 적절하지만 인물이 프레임 바깥으로 너무 밀려나 있고 어두워서 의상과 형태를 식별하기 어렵습니다." }, { "label": "D", "score": 3, "verdict_ko": "인물이 프레임에 너무 많이 들어와 구도감이 떨어지며, 의상이 검은색 재킷으로 렌더링되어 레퍼런스와 크게 어긋납니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L10B03.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "옥탑방 건물의 외벽 재질(붉은 벽돌)과 개구부 형태가 구조 기준 사진(회색 콘크리트, 우측 문과 좌측 창문 배치)을 따르지 않고 위치 기준 사진을 그대로 모방했습니다.", "fix_en": "Change the rooftop room's exterior walls to smooth grey concrete and adjust the placement of the door and window to match the structure look photograph." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the rooftop room's exterior walls to smooth grey concrete and adjust the placement of the door and window to match the structure look photograph.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+seed+엔티티 (복잡구조물 4택1: 무콘티 승·C=변형0)", "lane_policy": "ab_select_ready", "plate_select": { "candidates": { "A": "L10B01", "B": "L10B02", "C": "L10B03" }, "assigned": "L10B03", "choice": "Candidate C", "confident": true, "reason_ko": "\"수리영의 어깨 너머로 보이는, 어둠 속에서 반쯤 열린 채 멈춰 있는 옥탑방 현관 철문\"이라는 묘사에 정확히 부합하는 오버 더 숄더(Over-the-shoulder) 구도와 반쯤 열린 철문이 있는 서브공간입니다.", "kept": "L10B03" } }, "S12sh6::variants": { "author_fp": "90af9ed4064fdc50", "author": { "variants": [ { "approach_ko": "바닥 높이의 밀도 높은 근접 촬영으로 축 늘어진 자세와 어깨·쇄골 상처를 직접 강조한다.", "prompt_en": "Photograph her from near floor height in a tight, intimate medium close view, framing the dropped head, slack torso, collapsed hand with the crumpled photograph resting on its support, and the torn shoulder and clavicle. Use a restrained natural perspective with shallow depth, letting the curtain and blood-marked floor fall softly out of focus around her. Shape the dim stand-lamp and television illumination into subdued side light, preserving heavy shadow while retaining realistic detail in the wound, face, and blood." }, { "approach_ko": "커튼을 전경 프레임으로 삼은 거리감 있는 와이드 쇼트로 방의 핏자국과 고립된 시신을 함께 보여준다.", "prompt_en": "Use a static, distanced wide composition with the curtain occupying a strong foreground edge and her collapsed seated body isolated deeper in the room. Keep the bloodstains, footprints, pooled blood, and surrounding floor legible as a visual path toward her, while allowing the compact interior to press around the figure. Hold broad depth and understated contrast, with the stand lamp and television providing uneven low-level night illumination rather than spotlighting her." } ] }, "reused": false }, "S12sh6": { "input_fingerprint": "b133b40ad1ff75a9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흐르는 참혹한 상처를 입은 강민숙이 고개를 떨구고 바닥에 앉아 있는 모습.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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 motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, and one hand tightly clutches a crumpled photograph.\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): Minsuk is dead, seated motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, her shoulder and clavicle torn open and blood pooled and spattered around the room. Because she cannot grasp it later, one hand already tightly clutches a crumpled photograph.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흐르는 참혹한 상처를 입은 강민숙이 고개를 떨구고 바닥에 앉아 있는 모습.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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 motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, and one hand tightly clutches a crumpled photograph.\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): Minsuk is dead, seated motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, her shoulder and clavicle torn open and blood pooled and spattered around the room. Because she cannot grasp it later, one hand already tightly clutches a crumpled photograph.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her from near floor height in a tight, intimate medium close view, framing the dropped head, slack torso, collapsed hand with the crumpled photograph resting on its support, and the torn shoulder and clavicle. Use a restrained natural perspective with shallow depth, letting the curtain and blood-marked floor fall softly out of focus around her. Shape the dim stand-lamp and television illumination into subdued side light, preserving heavy shadow while retaining realistic detail in the wound, face, and blood.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흐르는 참혹한 상처를 입은 강민숙이 고개를 떨구고 바닥에 앉아 있는 모습.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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 motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, and one hand tightly clutches a crumpled photograph.\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): Minsuk is dead, seated motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, her shoulder and clavicle torn open and blood pooled and spattered around the room. Because she cannot grasp it later, one hand already tightly clutches a crumpled photograph.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a static, distanced wide composition with the curtain occupying a strong foreground edge and her collapsed seated body isolated deeper in the room. Keep the bloodstains, footprints, pooled blood, and surrounding floor legible as a visual path toward her, while allowing the compact interior to press around the figure. Hold broad depth and understated contrast, with the stand lamp and television providing uneven low-level night illumination rather than spotlighting her.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흐르는 참혹한 상처를 입은 강민숙이 고개를 떨구고 바닥에 앉아 있는 모습.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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 motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, and one hand tightly clutches a crumpled photograph.\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): Minsuk is dead, seated motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, her shoulder and clavicle torn open and blood pooled and spattered around the room. Because she cannot grasp it later, one hand already tightly clutches a crumpled photograph.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her from near floor height in a tight, intimate medium close view, framing the dropped head, slack torso, collapsed hand with the crumpled photograph resting on its support, and the torn shoulder and clavicle. Use a restrained natural perspective with shallow depth, letting the curtain and blood-marked floor fall softly out of focus around her. Shape the dim stand-lamp and television illumination into subdued side light, preserving heavy shadow while retaining realistic detail in the wound, face, and blood.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흐르는 참혹한 상처를 입은 강민숙이 고개를 떨구고 바닥에 앉아 있는 모습.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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 motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, and one hand tightly clutches a crumpled photograph.\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): Minsuk is dead, seated motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, her shoulder and clavicle torn open and blood pooled and spattered around the room. Because she cannot grasp it later, one hand already tightly clutches a crumpled photograph.\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 (한국인 여성, 40대 초반, 짙은색 머리, 평범한 중년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a static, distanced wide composition with the curtain occupying a strong foreground edge and her collapsed seated body isolated deeper in the room. Keep the bloodstains, footprints, pooled blood, and surrounding floor legible as a visual path toward her, while allowing the compact interior to press around the figure. Hold broad depth and understated contrast, with the stand lamp and television providing uneven low-level night illumination rather than spotlighting her.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S12sh6.png" }, { "label": "CHARACTER REFERENCE — 강민숙: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S12sh6.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "D", "ranking": [ "D", "C", "B", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "치명적인 상처가 누락되었으나, 레퍼런스와 일치하는 장소 구조, 조명(TV와 스탠드), 고개를 떨구고 중력에 순응하는 시신의 자세를 가장 잘 구현함." }, { "label": "C", "score": 5, "verdict_ko": "상처 묘사가 없고 고개를 옆으로 기대어 '고개를 떨군' 자세 지침을 위반했으며, 손에 쥔 사진의 형태가 다소 작위적임." }, { "label": "B", "score": 4, "verdict_ko": "고개를 뒤로 젖히고 있어 지정된 시신의 자세 지침을 정면으로 위반했으며, 요구된 상처 부위가 보이지 않음." }, { "label": "A", "score": 3, "verdict_ko": "사진을 들고 바라보는 등 근육에 힘이 들어간 산 사람의 자세를 취하고 있으며, 혈흔과 발자국이 완전히 누락됨." } ] }, "forward_normalized": { "winner": "D", "ranking": [ "D", "C", "B", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "치명적인 상처가 누락되었으나, 레퍼런스와 일치하는 장소 구조, 조명(TV와 스탠드), 고개를 떨구고 중력에 순응하는 시신의 자세를 가장 잘 구현함." }, { "label": "C", "score": 5, "verdict_ko": "상처 묘사가 없고 고개를 옆으로 기대어 '고개를 떨군' 자세 지침을 위반했으며, 손에 쥔 사진의 형태가 다소 작위적임." }, { "label": "B", "score": 4, "verdict_ko": "고개를 뒤로 젖히고 있어 지정된 시신의 자세 지침을 정면으로 위반했으며, 요구된 상처 부위가 보이지 않음." }, { "label": "A", "score": 3, "verdict_ko": "사진을 들고 바라보는 등 근육에 힘이 들어간 산 사람의 자세를 취하고 있으며, 혈흔과 발자국이 완전히 누락됨." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "D", "score": 6, "verdict_ko": "고개를 떨구고 축 늘어진 시신의 자세를 가장 잘 구현하고 중력 규칙을 준수했으나, 참혹한 상처와 핏자국이 묘사되지 않았습니다." }, { "label": "B", "score": 5, "verdict_ko": "바닥의 발자국은 잘 나타났으나, 고개를 앞으로 떨구지 않고 벽에 기대어 있으며 역시 상처 묘사가 누락되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "고개를 숙인 모습과 발자국은 있으나, 살아있는 듯한 능동적인 자세이며 사진을 든 손이 허공에 들려 있어 중력 규칙을 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "고개를 뒤로 젖히고 있어 지시문과 상반되며, 인물이 너무 멀리 배치되었고 상처 묘사도 없습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "고개를 떨구고 축 늘어진 시신의 자세를 가장 잘 구현하고 중력 규칙을 준수했으나, 참혹한 상처와 핏자국이 묘사되지 않았습니다." }, { "label": "C", "score": 5, "verdict_ko": "바닥의 발자국은 잘 나타났으나, 고개를 앞으로 떨구지 않고 벽에 기대어 있으며 역시 상처 묘사가 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "고개를 숙인 모습과 발자국은 있으나, 살아있는 듯한 능동적인 자세이며 사진을 든 손이 허공에 들려 있어 중력 규칙을 위반했습니다." }, { "label": "B", "score": 3, "verdict_ko": "고개를 뒤로 젖히고 있어 지시문과 상반되며, 인물이 너무 멀리 배치되었고 상처 묘사도 없습니다." } ] }, "combined": { "totals": { "A": 9, "B": 7, "C": 10, "D": 11 }, "ranking": [ "D", "C", "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 9, "B": 7, "C": 10, "D": 11 }, "selected": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "치명적인 상처가 누락되었으나, 레퍼런스와 일치하는 장소 구조, 조명(TV와 스탠드), 고개를 떨구고 중력에 순응하는 시신의 자세를 가장 잘 구현함." }, { "label": "C", "score": 5, "verdict_ko": "상처 묘사가 없고 고개를 옆으로 기대어 '고개를 떨군' 자세 지침을 위반했으며, 손에 쥔 사진의 형태가 다소 작위적임." }, { "label": "B", "score": 4, "verdict_ko": "고개를 뒤로 젖히고 있어 지정된 시신의 자세 지침을 정면으로 위반했으며, 요구된 상처 부위가 보이지 않음." }, { "label": "A", "score": 3, "verdict_ko": "사진을 들고 바라보는 등 근육에 힘이 들어간 산 사람의 자세를 취하고 있으며, 혈흔과 발자국이 완전히 누락됨." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "강민숙의 어깨와 쇄골 부위가 뜯겨져 피가 흐르는 심각한 상처가 묘사되지 않았고 주변에 고인 피도 없습니다.", "fix_en": "Add a severe, torn-open flesh wound with heavy bleeding to the character's shoulder and clavicle, and add pooled and spattered blood on the floor directly around her body." }, { "issue_ko": "의식이 없는 상태임에도 오른쪽 손으로 왼쪽 팔을 잡고 있어 근육에 힘이 들어가 있으며 상체가 꼿꼿합니다.", "fix_en": "Make the character's torso slump completely slack forward, and drop her right arm so it lies entirely limp on the floor, removing the muscular effort of gripping her other arm." }, { "issue_ko": "인물이 지시된 대로 커튼 뒤가 아니라 커튼 앞 문지방에 앉아 있습니다.", "fix_en": "Move the character backwards so that her seated body is positioned entirely behind the curtain inside the room." }, { "issue_ko": "오른쪽 침실 바닥에 놓인 종이 상자 측면에 금지된 텍스트가 적혀 있습니다.", "fix_en": "Remove all printed text and logos from the surface of the cardboard box in the right-hand bedroom." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add a severe, torn-open flesh wound with heavy bleeding to the character's shoulder and clavicle, and add pooled and spattered blood on the floor directly around her body.\n- Make the character's torso slump completely slack forward, and drop her right arm so it lies entirely limp on the floor, removing the muscular effort of gripping her other arm.\n- Move the character backwards so that her seated body is positioned entirely behind the curtain inside the room.\n- Remove all printed text and logos from the surface of the cardboard box in the right-hand bedroom.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+엔티티 (4택1: 무콘티 승·D=변형1)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B06", "choice": "Candidate F", "confident": false, "reason_ko": "샷 텍스트에서 '바닥에 앉아 있는 모습'이라고만 언급할 뿐 아파트 내의 특정 서브 공간(거실, 방, 화장실 등)을 명확히 지시하는 단서가 없으므로 현재 할당된 후보를 유지합니다.", "kept": "L04B06" } }, "S12sh9::variants": { "author_fp": "58ec8e23e00653c9", "author": { "variants": [ { "approach_ko": "손과 구겨진 사진을 정면에 가깝게 내려다보는 절제된 초근접 인서트로 물증의 충격을 강조한다.", "prompt_en": "Photograph this as a restrained extreme close insert from nearly overhead, framing the clenched hand and crumpled photograph with very little surrounding floor. Use a crisp, forensic plane of focus across the fingers and crushed paper folds, with the blood-marked background falling away softly. Let the stand lamp provide a dim warm side wash while the television adds a faint cool contamination to the shadows, emphasizing texture without revealing any new information on the photograph." }, { "approach_ko": "바닥 높이의 비스듬한 시점과 얕은 초점으로 주먹의 긴장과 사진의 구김을 촉각적으로 포착한다.", "prompt_en": "Place the camera at floor level in a tight oblique close-up, looking across the collapsed hand so the curled fingers dominate the near foreground and the crumpled photograph emerges between them. Use very shallow focus concentrated on the gripping fingertips and the nearest paper crease, allowing the rest to dissolve into dim, uneasy depth. Shape the existing stand-lamp and television light as opposing warm and cool edge tones, creating a tactile, intimate image with heavy shadow and no added visual elements." } ] }, "reused": false }, "S12sh9": { "input_fingerprint": "032bb9f7ab2fef1a", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 죽은 강민숙의 손안에 꽉 쥐어진 구겨진 사진 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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): Minsuk remains dead, seated motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, and one hand tightly clutches a crumpled photograph, amid the blood and footprints.\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 죽은 강민숙의 손안에 꽉 쥐어진 구겨진 사진 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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): Minsuk remains dead, seated motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, and one hand tightly clutches a crumpled photograph, amid the blood and footprints.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph this as a restrained extreme close insert from nearly overhead, framing the clenched hand and crumpled photograph with very little surrounding floor. Use a crisp, forensic plane of focus across the fingers and crushed paper folds, with the blood-marked background falling away softly. Let the stand lamp provide a dim warm side wash while the television adds a faint cool contamination to the shadows, emphasizing texture without revealing any new information on the photograph.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 죽은 강민숙의 손안에 꽉 쥐어진 구겨진 사진 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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): Minsuk remains dead, seated motionless on the floor behind the curtain with her torso slack and her head dropped forward; her limbs remain in their collapsed seated positions, and one hand tightly clutches a crumpled photograph, amid the blood and footprints.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera at floor level in a tight oblique close-up, looking across the collapsed hand so the curled fingers dominate the near foreground and the crumpled photograph emerges between them. Use very shallow focus concentrated on the gripping fingertips and the nearest paper crease, allowing the rest to dissolve into dim, uneasy depth. Shape the existing stand-lamp and television light as opposing warm and cool edge tones, creating a tactile, intimate image with heavy shadow and no added visual elements.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요청된 클로즈업 샷과 '꽉 쥐어진 구겨진 사진'이라는 핵심 지시를 매우 충실히 구현했으며, 바닥의 혈흔과 커튼 등 디테일도 잘 반영되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "사진을 '꽉 쥐고 있는' 모습이 아니라 손가락 위에 얹혀 있는 듯한 부자연스러운 연출로 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" } ], "critique": { "issues": [ { "issue_ko": "손등이 위를 향한(손바닥이 아래로 향한) 오른손임에도 불구하고, 종이를 감싸 쥔 손가락들의 손톱이 카메라 쪽으로 위를 향해 노출되어 있어 해부학적으로 불가능한 구조입니다.", "fix_en": "Correct the right hand's anatomy so the fingers curl naturally downward around the crumpled paper, hiding the fingernails as they would in a palm-down fist." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Correct the right hand's anatomy so the fingers curl naturally downward around the crumpled paper, hiding the fingernails as they would in a palm-down fist.\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.", "ref_mode": "플레이트만 (배경 전용)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B06", "choice": "Candidate F", "confident": true, "reason_ko": "원문은 죽은 사람의 손을 비추는 클로즈업 숏으로 구체적인 공간적 배경이 명시되어 있지 않으므로, 명확한 변경 근거가 없어 현재 할당된 이미지를 유지합니다.", "kept": "L04B06" } }, "S12sh12::variants": { "author_fp": "48c04fa7d9d29281", "author": { "variants": [ { "approach_ko": "침대 정면의 눈높이 대칭 구도로 붉은 원형 표식과 방의 정적을 냉정하게 기록한다.", "prompt_en": "Photograph the room in a restrained, straight-on medium-wide view at eye level, with the bed anchoring the lower frame and the enormous red circle dominating the wall above it. Use near-symmetrical composition and deep focus to make the space feel clinically still, while the stand lamp and television create subdued, uneven pools of night interior light." }, { "approach_ko": "침대 가까이의 낮은 사선 시점과 압축된 원근으로 붉은 원이 벽면을 위압적으로 점유하게 만든다.", "prompt_en": "Place the camera low and close beside the bed, looking obliquely upward so the bed forms a strong foreground plane and the huge red circle looms across the upper frame. Use a compressed, claustrophobic lens feel with shallow falloff toward the rest of the room, letting the stand lamp provide soft lateral modeling while the television adds a faint fluctuating ambient cast." } ] }, "reused": false }, "S12sh12": { "input_fingerprint": "e0af2b059202b614", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 침대 위 벽면에 거대한 붉은색 원형 표식이 나타난 모습.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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): Minsuk’s corpse remains seated behind the curtain with its wounds and the room’s blood traces unchanged, but Suri-young has removed the crumpled photograph from her hand. Minsuk’s wrist mark is now red, and a large red circle remains on the wall above the bed.\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 침대 위 벽면에 거대한 붉은색 원형 표식이 나타난 모습.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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): Minsuk’s corpse remains seated behind the curtain with its wounds and the room’s blood traces unchanged, but Suri-young has removed the crumpled photograph from her hand. Minsuk’s wrist mark is now red, and a large red circle remains on the wall above the bed.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the room in a restrained, straight-on medium-wide view at eye level, with the bed anchoring the lower frame and the enormous red circle dominating the wall above it. Use near-symmetrical composition and deep focus to make the space feel clinically still, while the stand lamp and television create subdued, uneven pools of night interior light.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with a stand lamp and television on.\n\nSHOT TEXT (authoritative, Korean): 침대 위 벽면에 거대한 붉은색 원형 표식이 나타난 모습.\n\nLOCATION (lock): A compact rooftop apartment containing a living room, dining table, main bedroom, and a separate bedroom with a curtained window. Bloodstains and footprints cover the latter room and lead behind the curtain. 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): Minsuk’s corpse remains seated behind the curtain with its wounds and the room’s blood traces unchanged, but Suri-young has removed the crumpled photograph from her hand. Minsuk’s wrist mark is now red, and a large red circle remains on the wall above the bed.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera low and close beside the bed, looking obliquely upward so the bed forms a strong foreground plane and the huge red circle looms across the upper frame. Use a compressed, claustrophobic lens feel with shallow falloff toward the rest of the room, letting the stand lamp provide soft lateral modeling while the television adds a faint fluctuating ambient cast.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 8, "B": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 붉은색 원형 표식, TV와 스탠드 조명, 커튼으로 향하는 발자국 등 모든 공간적 디테일을 프롬프트에 맞게 정확히 구현했으며, 인물 미등장 조건도 잘 지켰습니다." }, { "label": "B", "score": 4, "verdict_ko": "지문에서 인물이나 신체 부위가 보이지 않아야 한다고 명시했으나 시신이 프레임에 노출되었고, 발자국 방향이 커튼 쪽이 아닌 반대쪽을 향하고 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" } ], "critique": { "issues": [ { "issue_ko": "바닥의 핏빛 발자국 방향이 커튼 뒤로 향하지 않고 반대로 카메라(앞쪽)를 향해 걸어 나온 방향으로 찍혀 있습니다.", "fix_en": "Reverse the direction of the bloody footprints on the floor so the toes point toward the curtain." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Reverse the direction of the bloody footprints on the floor so the toes point toward the curtain.\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.", "ref_mode": "플레이트만 (배경 전용)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B06", "choice": "Candidate F", "confident": true, "reason_ko": "원문의 '침대 위 벽면'이라는 묘사에 따라 우측에 침대와 그 위쪽 벽면이 보이는 이 후보가 해당 숏의 배경으로 가장 적합합니다.", "kept": "L04B06" } }, "S13sh1::variants": { "author_fp": "c9b7cb92be0c6ed9", "author": { "variants": [ { "approach_ko": "계단 아래의 낮은 근접 시점과 넓은 화각으로 공중에 뜬 발, 전신의 낙하 운동, 공포의 얼굴을 강렬하게 포착한다.", "prompt_en": "Photograph from low near the foot of the staircase with a close, wide-angle perspective, keeping her entire body in frame while the descending steps surge diagonally behind her. Freeze the instant of suspension crisply, with the lifted foot clearly separated from the stair and the crumpled photograph readable as a physical object in her grip. Use hard lateral night light to catch her terrified expression and the strain of motion, allowing the surrounding courtyard to fall into restrained shadow." }, { "approach_ko": "안마당 쪽의 떨어진 시점에서 망원 압축으로 계단 구조에 갇힌 듯한 전신과 다급한 움직임을 관찰한다.", "prompt_en": "Observe from farther across the courtyard and vehicle-access area with a compressed telephoto feel, framing her full body tightly within the descending geometry of the exterior staircase. Layer the stair structure around her to create a trapped, urgent composition while preserving a clear view of the airborne foot, frightened face, and carried photograph. Shape the night illumination as a narrow pool across her path, with softer falloff into the surrounding exterior." } ] }, "reused": false }, "S13sh1": { "input_fingerprint": "bad9adaf8c3734e7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 공포에 질린 얼굴의 수리영이 옥탑방 외부 계단을 뛰어내려오는 도중, 한 발이 허공에 뜬 mid-action 전신.\n\nLOCATION (lock): The exterior approach to a rooftop dwelling, including the villa courtyard, the descending exterior staircase, and the nearby vehicle-access area. 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): Suri-young flees immediately after taking the crumpled photograph from Minsuk’s hand, so she still carries it as she runs down the stairs.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 공포에 질린 얼굴의 수리영이 옥탑방 외부 계단을 뛰어내려오는 도중, 한 발이 허공에 뜬 mid-action 전신.\n\nLOCATION (lock): The exterior approach to a rooftop dwelling, including the villa courtyard, the descending exterior staircase, and the nearby vehicle-access area. 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): Suri-young flees immediately after taking the crumpled photograph from Minsuk’s hand, so she still carries it as she runs down the stairs.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from low near the foot of the staircase with a close, wide-angle perspective, keeping her entire body in frame while the descending steps surge diagonally behind her. Freeze the instant of suspension crisply, with the lifted foot clearly separated from the stair and the crumpled photograph readable as a physical object in her grip. Use hard lateral night light to catch her terrified expression and the strain of motion, allowing the surrounding courtyard to fall into restrained shadow.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 공포에 질린 얼굴의 수리영이 옥탑방 외부 계단을 뛰어내려오는 도중, 한 발이 허공에 뜬 mid-action 전신.\n\nLOCATION (lock): The exterior approach to a rooftop dwelling, including the villa courtyard, the descending exterior staircase, and the nearby vehicle-access area. 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): Suri-young flees immediately after taking the crumpled photograph from Minsuk’s hand, so she still carries it as she runs down the stairs.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nObserve from farther across the courtyard and vehicle-access area with a compressed telephoto feel, framing her full body tightly within the descending geometry of the exterior staircase. Layer the stair structure around her to create a trapped, urgent composition while preserving a clear view of the airborne foot, frightened face, and carried photograph. Shape the night illumination as a narrow pool across her path, with softer falloff into the surrounding exterior.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 7, "A": 5 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "스토리보드의 정면 구도, 캐릭터의 지정된 의상, 공중 체공 동작 및 구겨진 사진 소지까지 모든 핵심 지시를 매우 충실하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "공중에 뜬 발과 공포에 질린 표정, 사진 소지 상태는 좋으나, 레퍼런스의 파란 점퍼와 청바지 의상을 누락했고 카메라 구도가 스토리보드와 어긋납니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S13sh1.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영이 들고 있는 사진이 구겨져 있지 않으며, 오른손 손가락과 사진이 기형적으로 융합되어 있습니다.", "fix_en": "Restore the right hand to normal anatomy and make it hold a clearly crumpled photograph." }, { "issue_ko": "캐릭터 레퍼런스 이미지에서 수리영이 착용하고 있는 헬멧과 올리브색 크로스백(힙색)이 누락되었습니다.", "fix_en": "Add the black helmet and olive-green crossbody bag from the character reference to Suri-young." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the right hand to normal anatomy and make it hold a clearly crumpled photograph.\n- Add the black helmet and olive-green crossbody bag from the character reference to Suri-young.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S13sh3::variants": { "author_fp": "1c975a2482abb307", "author": { "variants": [ { "approach_ko": "두 사람을 같은 비중으로 담는 밀착된 눈높이 측면 투숏으로, 대치와 단절감을 동시에 강조한다.", "prompt_en": "Photograph the moment as a tight eye-level upper-body two-shot, holding both women in opposing near-profile with their faces and the shoulder grip clearly readable in the same plane. Use restrained shallow focus to isolate their confrontation while the nighttime exterior recedes softly, emphasizing the painful contrast between Hye-su’s urgent physical contact and Suri-young’s unfocused stillness." }, { "approach_ko": "혜수의 어깨 너머로 수리영의 멍한 얼굴을 압도적으로 포착하는 반응 중심의 구도다.", "prompt_en": "Frame from close behind Hye-su in an intimate over-the-shoulder upper-body composition, using her foreground presence and gripping hand to enclose Suri-young within the frame. Keep Suri-young’s face sharply dominant and let Hye-su fall slightly softer, with compressed depth and subdued directional night light concentrating attention on Suri-young’s unfixed gaze and rigid posture." } ] }, "reused": false }, "S13sh3": { "input_fingerprint": "4786e2fa8ab711a6", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수가 초점을 잃고 서 있는 수리영의 어깨를 쥔 채 마주 보는 상체 구도.\n\nLOCATION (lock): The exterior approach to a rooftop dwelling, including the villa courtyard, the descending exterior staircase, and the nearby vehicle-access area. 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): Suri-young remains stunned outside and still has the crumpled photograph she removed before fleeing 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수가 초점을 잃고 서 있는 수리영의 어깨를 쥔 채 마주 보는 상체 구도.\n\nLOCATION (lock): The exterior approach to a rooftop dwelling, including the villa courtyard, the descending exterior staircase, and the nearby vehicle-access area. 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): Suri-young remains stunned outside and still has the crumpled photograph she removed before fleeing 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a tight eye-level upper-body two-shot, holding both women in opposing near-profile with their faces and the shoulder grip clearly readable in the same plane. Use restrained shallow focus to isolate their confrontation while the nighttime exterior recedes softly, emphasizing the painful contrast between Hye-su’s urgent physical contact and Suri-young’s unfocused stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수가 초점을 잃고 서 있는 수리영의 어깨를 쥔 채 마주 보는 상체 구도.\n\nLOCATION (lock): The exterior approach to a rooftop dwelling, including the villa courtyard, the descending exterior staircase, and the nearby vehicle-access area. 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): Suri-young remains stunned outside and still has the crumpled photograph she removed before fleeing 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame from close behind Hye-su in an intimate over-the-shoulder upper-body composition, using her foreground presence and gripping hand to enclose Suri-young within the frame. Keep Suri-young’s face sharply dominant and let Hye-su fall slightly softer, with compressed depth and subdued directional night light concentrating attention on Suri-young’s unfixed gaze and rigid posture.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 8, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "프레임 밖으로 손이 제외되어 사진을 억지로 노출하지 않은 점이 훌륭하며, 인물의 멍한 표정과 상체 구도를 매우 잘 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "수리영의 팔과 혜수의 손 사이에 구겨진 사진이 부자연스럽게 합성되어 물리적으로 불가능한 묘사가 발생했습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S13sh3.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "혜수의 오른쪽 소매에 금지된 텍스트가 적혀 있습니다.", "fix_en": "Remove the text from the band on Hye-su's right sleeve." }, { "issue_ko": "수리영이 캐릭터 레퍼런스와 달리 긴 머리를 하고 있으며, 검은색 헬멧과 가슴에 메는 초록색 가방이 누락되었습니다.", "fix_en": "Remove Suri-young's long hair, putting the black helmet on her head and adding the green sling bag across her chest to match her reference." }, { "issue_ko": "혜수가 캐릭터 레퍼런스와 달리 포니테일을 하고 있으며 남색 모자가 없습니다.", "fix_en": "Change Hye-su's ponytail to loose shoulder-length hair and place the navy cap from her reference on her head." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the text from the band on Hye-su's right sleeve.\n- Remove Suri-young's long hair, putting the black helmet on her head and adding the green sling bag across her chest to match her reference.\n- Change Hye-su's ponytail to loose shoulder-length hair and place the navy cap from her reference on her head.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S14sh3::variants": { "author_fp": "ec6f9f0da9884439", "author": { "variants": [ { "approach_ko": "정면 와이드 고정 구도로 수리영의 정지와 기이할 만큼 완벽한 방의 질서를 동시에 강조한다.", "prompt_en": "Photograph the moment as a restrained, eye-level wide tableau, with the camera square to the room and Suriyoung held near the compositional center. Use deep focus and clean geometric lines so the precise arrangement of the bedroom surrounds her rigid stillness with oppressive symmetry. Preserve the established nighttime illumination, allowing restrained contrast and a slightly underexposed facial complexion to convey her shock through performance rather than altered anatomy." }, { "approach_ko": "얼굴과 경직된 자세에 밀착한 비스듬한 클로즈업으로 충격을 내면화하고 정돈된 방은 압축된 배경으로 남긴다.", "prompt_en": "Use an intimate, slightly off-axis close framing from around chest height, concentrating on Suriyoung’s pallid face, fixed gaze, and arrested posture. Let a longer-lens feel compress the precisely ordered room behind her into layered, softly receding planes while retaining enough recognizable structure to make its unnatural neatness inescapable. Keep the established nighttime light natural to the location, with gentle directional modeling across her face and shallow focus isolating her stunned performance." } ] }, "reused": false }, "S14sh3": { "input_fingerprint": "92e844c08ceb9113", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핏자국 하나 없이 깨끗하게 정돈된 방 한가운데 서서 파랗게 질린 얼굴로 멈춘 수리영.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 bedroom layout, fixed wall and floor materials, curtain placement, and nighttime lighting. Preserve the young woman's clothing from this continuous night sequence. Exclude the dead woman, all blood, torn flesh, and every trace of the earlier attack; the room must now appear unnaturally clean and orderly.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 room has been unnaturally reset: the corpse, blood, bloody footprints, wall circle, and other visible evidence have all disappeared.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핏자국 하나 없이 깨끗하게 정돈된 방 한가운데 서서 파랗게 질린 얼굴로 멈춘 수리영.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 bedroom layout, fixed wall and floor materials, curtain placement, and nighttime lighting. Preserve the young woman's clothing from this continuous night sequence. Exclude the dead woman, all blood, torn flesh, and every trace of the earlier attack; the room must now appear unnaturally clean and orderly.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 room has been unnaturally reset: the corpse, blood, bloody footprints, wall circle, and other visible evidence have all disappeared.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a restrained, eye-level wide tableau, with the camera square to the room and Suriyoung held near the compositional center. Use deep focus and clean geometric lines so the precise arrangement of the bedroom surrounds her rigid stillness with oppressive symmetry. Preserve the established nighttime illumination, allowing restrained contrast and a slightly underexposed facial complexion to convey her shock through performance rather than altered anatomy.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핏자국 하나 없이 깨끗하게 정돈된 방 한가운데 서서 파랗게 질린 얼굴로 멈춘 수리영.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 bedroom layout, fixed wall and floor materials, curtain placement, and nighttime lighting. Preserve the young woman's clothing from this continuous night sequence. Exclude the dead woman, all blood, torn flesh, and every trace of the earlier attack; the room must now appear unnaturally clean and orderly.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 room has been unnaturally reset: the corpse, blood, bloody footprints, wall circle, and other visible evidence have all disappeared.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate, slightly off-axis close framing from around chest height, concentrating on Suriyoung’s pallid face, fixed gaze, and arrested posture. Let a longer-lens feel compress the precisely ordered room behind her into layered, softly receding planes while retaining enough recognizable structure to make its unnatural neatness inescapable. Keep the established nighttime light natural to the location, with gentle directional modeling across her face and shallow focus isolating her stunned performance.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 7, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "창백하게 질린 얼굴과 깨끗하게 정돈된 방 한가운데 선 모습이 잘 연출되었으나, 침대의 위치가 기준 이미지와 반대로 뒤집혀 있습니다." }, { "label": "A", "score": 3, "verdict_ko": "인물의 표정이 잘 보이지 않을 정도로 프레이밍이 넓고, 방과 복도의 구조가 무대 세트처럼 왜곡되는 치명적인 구조적 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/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/scene/recipe/S12sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영이 캐릭터 레퍼런스 이미지에서 착용하고 있는 검은색 자전거 헬멧을 쓰고 있지 않습니다.", "fix_en": "Add the black bicycle helmet onto Suriyoung's head, matching the character reference exactly." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the black bicycle helmet onto Suriyoung's head, matching the character reference exactly.\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.", "ref_mode": "prev+엔티티" }, "S14sh4::variants": { "author_fp": "f59d7a3d8424488c", "author": { "variants": [ { "approach_ko": "바닥 높이의 밀착된 시점으로 커튼 너머의 깨끗한 빈 공간을 불길한 부재감으로 강조한다.", "prompt_en": "Photograph from floor level at close range, looking just beyond the curtain so its edge forms a soft foreground veil around the sharply rendered, immaculate empty floor. Use a restrained natural-perspective lens feel and shallow, controlled depth, with low raking interior light revealing the spotless surface and exact arrangement while the untouched negative space carries the tension." }, { "approach_ko": "높은 고정 시점의 절제된 와이드 구도로 정돈된 침실과 비어 있는 커튼 뒤 바닥의 기하학적 대비를 보여준다.", "prompt_en": "Use a high, static wide composition that observes the precisely arranged bedroom and the empty floor behind the curtain as a cold geometric tableau. Keep the frame deep and evenly legible, with the curtain establishing a clear spatial boundary and subdued night-interior illumination emphasizing unnatural order, cleanliness, and absence." } ] }, "reused": false }, "S14sh4": { "input_fingerprint": "fe28b22a2f61c5d4", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 시신이 사라지고 빈 공간만 남은 깨끗한 커튼 뒤 바닥.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 floor behind the curtain remains completely clean and empty, with Minsuk’s seated corpse and all blood traces gone.\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 시신이 사라지고 빈 공간만 남은 깨끗한 커튼 뒤 바닥.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 floor behind the curtain remains completely clean and empty, with Minsuk’s seated corpse and all blood traces gone.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from floor level at close range, looking just beyond the curtain so its edge forms a soft foreground veil around the sharply rendered, immaculate empty floor. Use a restrained natural-perspective lens feel and shallow, controlled depth, with low raking interior light revealing the spotless surface and exact arrangement while the untouched negative space carries the tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 시신이 사라지고 빈 공간만 남은 깨끗한 커튼 뒤 바닥.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 floor behind the curtain remains completely clean and empty, with Minsuk’s seated corpse and all blood traces gone.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a high, static wide composition that observes the precisely arranged bedroom and the empty floor behind the curtain as a cold geometric tableau. Keep the frame deep and evenly legible, with the curtain establishing a clear spatial boundary and subdued night-interior illumination emphasizing unnatural order, cleanliness, and absence.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "프롬프트가 지시한 '커튼 뒤의 빈 바닥'이라는 세밀한 구도를 기준 이미지의 공간 내에서 충실하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "지정된 옥탑방 내부라는 공간 설정을 무시하고 고층 빌딩 외부에서 방을 들여다보는 잘못된 구도를 연출했습니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" } ], "critique": { "issues": [] }, "fix_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B07", "choice": "Candidate G", "confident": false, "reason_ko": "지문에서 '깨끗한 커튼 뒤 바닥'과 '빈 공간'을 요구하며, 로케이션 설명의 '비정상적으로 깨끗해진 침실' 설정과 가장 부합하는 정돈된 공간을 보여주는 후보가 G이므로 현재 할당을 유지합니다.", "kept": "L04B07" } }, "S14sh8::variants": { "author_fp": "9822ba4c866cc5bf", "author": { "variants": [ { "approach_ko": "눈높이의 밀도 높은 측면 투샷으로 두 얼굴과 어깨를 붙잡은 손의 긴장을 동시에 포착한다.", "prompt_en": "Photograph the moment as an intimate eye-level medium close two-shot from a slight side angle, keeping both faces and the contact at the shoulders clearly legible in the same plane. Use a gently compressed lens feel and shallow but controlled focus, with soft directional interior light shaping natural skin texture and restrained shadows; let the rigid stillness and the pressure of the hands carry the frame." }, { "approach_ko": "방 한쪽의 넓고 고정된 시점에서 두 인물을 정돈된 침실의 기하학 속에 작게 배치해 불길한 정적을 강조한다.", "prompt_en": "Use a wide, static view from one side of the bedroom, placing the two standing figures within the unnaturally precise geometry of the clean room rather than isolating them. Keep the foreground, their full posture, and the altered positions of the backpack and doll readable through layered depth, with restrained night-interior illumination and broad areas of quiet shadow creating an observational, unsettling stillness." } ] }, "reused": false }, "S14sh8": { "input_fingerprint": "8d56f698a1335858", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 혜수가 수리영의 어깨를 양손으로 감싸 쥔 채 서 있는 정지 순간.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 apartment remains clean of the corpse, blood, wall mark, and two teacups; Suri-young’s backpack and doll now sit in slightly altered positions in her 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 혜수가 수리영의 어깨를 양손으로 감싸 쥔 채 서 있는 정지 순간.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 apartment remains clean of the corpse, blood, wall mark, and two teacups; Suri-young’s backpack and doll now sit in slightly altered positions in her 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as an intimate eye-level medium close two-shot from a slight side angle, keeping both faces and the contact at the shoulders clearly legible in the same plane. Use a gently compressed lens feel and shallow but controlled focus, with soft directional interior light shaping natural skin texture and restrained shadows; let the rigid stillness and the pressure of the hands carry the frame.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 혜수가 수리영의 어깨를 양손으로 감싸 쥔 채 서 있는 정지 순간.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 apartment remains clean of the corpse, blood, wall mark, and two teacups; Suri-young’s backpack and doll now sit in slightly altered positions in her 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wide, static view from one side of the bedroom, placing the two standing figures within the unnaturally precise geometry of the clean room rather than isolating them. Keep the foreground, their full posture, and the altered positions of the backpack and doll readable through layered depth, with restrained night-interior illumination and broad areas of quiet shadow creating an observational, unsettling stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 혜수가 수리영의 어깨를 양손으로 감싸 쥔 채 서 있는 정지 순간.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 apartment remains clean of the corpse, blood, wall mark, and two teacups; Suri-young’s backpack and doll now sit in slightly altered positions in her 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as an intimate eye-level medium close two-shot from a slight side angle, keeping both faces and the contact at the shoulders clearly legible in the same plane. Use a gently compressed lens feel and shallow but controlled focus, with soft directional interior light shaping natural skin texture and restrained shadows; let the rigid stillness and the pressure of the hands carry the frame.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 혜수가 수리영의 어깨를 양손으로 감싸 쥔 채 서 있는 정지 순간.\n\nLOCATION (lock): A compact rooftop apartment with an iron-door entry, living and dining area, and a small bedroom. The bedroom is now unnaturally clean and precisely arranged, with no visible blood or body. 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 apartment remains clean of the corpse, blood, wall mark, and two teacups; Suri-young’s backpack and doll now sit in slightly altered positions in her 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wide, static view from one side of the bedroom, placing the two standing figures within the unnaturally precise geometry of the clean room rather than isolating them. Keep the foreground, their full posture, and the altered positions of the backpack and doll readable through layered depth, with restrained night-interior illumination and broad areas of quiet shadow creating an observational, unsettling stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S14sh8.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 혜수: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S14sh8.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "C", "B", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "양손으로 어깨를 감싸 쥔 핵심 동작과 지정된 실내 아키텍처, 침실의 소품 배치를 완벽히 따랐으나 두 인물의 의상 디테일이 누락되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물들의 의상과 지정된 소품은 매우 정확히 묘사했으나, 양손이 아닌 한 손만 어깨에 올려 최우선 우선순위인 동작 지시를 위반했습니다." }, { "label": "C", "score": 5, "verdict_ko": "양손으로 어깨를 쥐는 동작은 충족했으나, 거실 베란다 창문 앞에 침대가 놓여 있는 등 레퍼런스의 배경 구조를 심각하게 왜곡했습니다." }, { "label": "D", "score": 3, "verdict_ko": "한 손만 올린 동작 오류에 더해 참조 사진과 전혀 일치하지 않는 대칭형의 낯선 방 구조를 그려내어 배경 지시를 완전히 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "C", "B", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "양손으로 어깨를 감싸 쥔 핵심 동작과 지정된 실내 아키텍처, 침실의 소품 배치를 완벽히 따랐으나 두 인물의 의상 디테일이 누락되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물들의 의상과 지정된 소품은 매우 정확히 묘사했으나, 양손이 아닌 한 손만 어깨에 올려 최우선 우선순위인 동작 지시를 위반했습니다." }, { "label": "C", "score": 5, "verdict_ko": "양손으로 어깨를 쥐는 동작은 충족했으나, 거실 베란다 창문 앞에 침대가 놓여 있는 등 레퍼런스의 배경 구조를 심각하게 왜곡했습니다." }, { "label": "D", "score": 3, "verdict_ko": "한 손만 올린 동작 오류에 더해 참조 사진과 전혀 일치하지 않는 대칭형의 낯선 방 구조를 그려내어 배경 지시를 완전히 위반했습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "양손으로 어깨를 감싸 쥔 동작과 아파트 내부 구조를 가장 정확히 구현했으나 의상이 레퍼런스와 다릅니다." }, { "label": "B", "score": 4, "verdict_ko": "양손 동작은 맞으나 침대가 발코니가 있는 거실에 배치되는 등 공간 구조 왜곡이 심각합니다." }, { "label": "C", "score": 3, "verdict_ko": "양손이 아닌 한 손만 어깨에 올려 핵심 동작 지시를 위반했습니다." }, { "label": "A", "score": 2, "verdict_ko": "한 손만 어깨에 올렸으며, 방 양쪽에 문이 존재하는 등 공간 구조 오류가 큽니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "C", "B", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "양손으로 어깨를 감싸 쥔 동작과 아파트 내부 구조를 가장 정확히 구현했으나 의상이 레퍼런스와 다릅니다." }, { "label": "C", "score": 4, "verdict_ko": "양손 동작은 맞으나 침대가 발코니가 있는 거실에 배치되는 등 공간 구조 왜곡이 심각합니다." }, { "label": "B", "score": 3, "verdict_ko": "양손이 아닌 한 손만 어깨에 올려 핵심 동작 지시를 위반했습니다." }, { "label": "D", "score": 2, "verdict_ko": "한 손만 어깨에 올렸으며, 방 양쪽에 문이 존재하는 등 공간 구조 오류가 큽니다." } ] }, "combined": { "totals": { "A": 14, "B": 7, "C": 9, "D": 5 }, "ranking": [ "A", "C", "B", "D" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 7, "C": 9, "D": 5 }, "selected": "A", "ranking": [ "A", "C", "B", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "양손으로 어깨를 감싸 쥔 핵심 동작과 지정된 실내 아키텍처, 침실의 소품 배치를 완벽히 따랐으나 두 인물의 의상 디테일이 누락되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물들의 의상과 지정된 소품은 매우 정확히 묘사했으나, 양손이 아닌 한 손만 어깨에 올려 최우선 우선순위인 동작 지시를 위반했습니다." }, { "label": "C", "score": 5, "verdict_ko": "양손으로 어깨를 쥐는 동작은 충족했으나, 거실 베란다 창문 앞에 침대가 놓여 있는 등 레퍼런스의 배경 구조를 심각하게 왜곡했습니다." }, { "label": "D", "score": 3, "verdict_ko": "한 손만 올린 동작 오류에 더해 참조 사진과 전혀 일치하지 않는 대칭형의 낯선 방 구조를 그려내어 배경 지시를 완전히 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "혜수가 수리영의 어깨를 양손으로 감싸 쥐어야 한다는 텍스트 지시와 달리, 한 손으로만 어깨를 잡고 있습니다.", "fix_en": "Add Hyesu's other hand grasping Suri-young's far shoulder to show her holding with both hands." }, { "issue_ko": "두 인물 모두 레퍼런스 이미지에 지정된 본인들의 복장(혜수의 모자와 조끼, 수리영의 헬멧과 파란 재킷)을 착용하지 않았습니다.", "fix_en": "Dress Hyesu in her reference cap and uniform vest, and dress Suri-young in her reference helmet and blue jacket." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add Hyesu's other hand grasping Suri-young's far shoulder to show her holding with both hands.\n- Dress Hyesu in her reference cap and uniform vest, and dress Suri-young in her reference helmet and blue jacket.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·A=변형0)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B07", "choice": "Candidate G", "confident": false, "reason_ko": "지문에서 두 인물이 서 있는 구체적인 공간(거실, 침실, 주방 등)이 명시되지 않아 특정 서브공간을 확정할 수 없으므로, 규칙에 따라 기존에 할당된 후보를 유지합니다.", "kept": "L04B07" } }, "S15sh1::variants": { "author_fp": "1fa14554c0e4116b", "author": { "variants": [ { "approach_ko": "눈높이의 절제된 와이드 숏으로 인물의 고립감과 어두운 바다의 깊이를 강조한다.", "prompt_en": "Photograph the moment as a restrained eye-level wide shot, holding the seated figure relatively small within generous negative space and letting the dark sea recede across the background. Use a natural, moderately deep focus and quiet low-key night illumination, with subtle separation around the face and posture while the surrounding environment falls into dense shadow." }, { "approach_ko": "가까운 망원 인물 숏으로 멍한 표정과 정지된 자세를 압축해 포착한다.", "prompt_en": "Photograph the moment as an intimate, gently compressed portrait from near seated eye level, framing tightly enough for the vacant performance and motionless posture to dominate while retaining only a soft suggestion of the seaside setting behind her. Use shallow focus, restrained contrast, and soft directional night light across the face, allowing the background to dissolve into dark, subdued layers." } ] }, "reused": false }, "S15sh1": { "input_fingerprint": "f7e2185c75f7227e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다 앞 버스정류장 벤치에 수리영이 멍한 표정으로 앉아 있는 모습.\n\nLOCATION (lock): A coastal bus-stop area with seating facing the open sea, set along the shoreline road. 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): Suri-young still has her mobile phone, but Minsuk remains unreachable.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다 앞 버스정류장 벤치에 수리영이 멍한 표정으로 앉아 있는 모습.\n\nLOCATION (lock): A coastal bus-stop area with seating facing the open sea, set along the shoreline road. 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): Suri-young still has her mobile phone, but Minsuk remains unreachable.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a restrained eye-level wide shot, holding the seated figure relatively small within generous negative space and letting the dark sea recede across the background. Use a natural, moderately deep focus and quiet low-key night illumination, with subtle separation around the face and posture while the surrounding environment falls into dense shadow.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다 앞 버스정류장 벤치에 수리영이 멍한 표정으로 앉아 있는 모습.\n\nLOCATION (lock): A coastal bus-stop area with seating facing the open sea, set along the shoreline road. 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): Suri-young still has her mobile phone, but Minsuk remains unreachable.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as an intimate, gently compressed portrait from near seated eye level, framing tightly enough for the vacant performance and motionless posture to dominate while retaining only a soft suggestion of the seaside setting behind her. Use shallow focus, restrained contrast, and soft directional night light across the face, allowing the background to dissolve into dark, subdued layers.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "스토리보드가 지정한 카메라 구도와 앵글을 훌륭하게 구현했으며, 레퍼런스의 복장(헬멧, 파란색 재킷)과 소지품(휴대폰)을 정확히 반영했습니다." }, { "label": "B", "score": 3, "verdict_ko": "스토리보드의 와이드 샷 프레이밍과 카메라 위치를 완전히 무시했으며, 지정된 캐릭터의 필수 복장(헬멧 등)을 누락했습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S15sh1.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프롬프트와 스토리보드 스케치에 명시된 것과 달리, 벤치와 인물이 바다가 아닌 카메라(도로)를 향해 앉아 있습니다.", "fix_en": "Rotate the bench and the character 180 degrees so she sits with her back to the camera, facing the open sea." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Rotate the bench and the character 180 degrees so she sits with her back to the camera, facing the open sea.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S15sh4::variants": { "author_fp": "f402e2c1449771a0", "author": { "variants": [ { "approach_ko": "스토리보드 구도를 유지하며 낮은 추적 시점과 깊은 초점으로 공중에 뜬 발과 해안도로의 진행감을 선명하게 포착한다.", "prompt_en": "Honor the storyboard’s camera and figure placement with a low, trailing perspective and a moderately wide lens feel. Freeze Suri-young sharply at the decisive mid-stride instant, keeping the airborne foot, her receding back, and the carried phone clearly legible; use deep focus and strong road-leading geometry to drive the eye toward her path. Shape the night with cool coastal ambience and restrained directional roadside spill, preserving natural detail rather than reducing her to a silhouette." }, { "approach_ko": "스토리보드 배치를 지키되 압축된 망원 느낌과 얕은 초점, 절제된 움직임 잔상으로 멀어지는 달리기의 긴박함을 강조한다.", "prompt_en": "Preserve the storyboard’s staging while giving the image a compressed, distant lens feel, visually stacking Suri-young against the coastal roadway and sea-facing bus-stop area. Hold her back and airborne foot at the focal center with shallow depth, allowing a restrained trace of motion in her clothing and grounded leg while the mid-stride pose remains unmistakable. Use soft, low-key night illumination with a narrow rim of coastal ambient light to separate her natural human form from the darker surroundings." } ] }, "reused": false }, "S15sh4": { "input_fingerprint": "17c6920a538bf31d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영이 해안가 도로를 따라 한 발이 허공에 떠 있는 미드 스트라이드(mid-stride) 자세로 달리는 뒷모습.\n\nLOCATION (lock): A coastal bus-stop area with seating facing the open sea, set along the shoreline road. 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): Suri-young continues carrying her mobile phone as she runs away from the bus stop.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영이 해안가 도로를 따라 한 발이 허공에 떠 있는 미드 스트라이드(mid-stride) 자세로 달리는 뒷모습.\n\nLOCATION (lock): A coastal bus-stop area with seating facing the open sea, set along the shoreline road. 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): Suri-young continues carrying her mobile phone as she runs away from the bus stop.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nHonor the storyboard’s camera and figure placement with a low, trailing perspective and a moderately wide lens feel. Freeze Suri-young sharply at the decisive mid-stride instant, keeping the airborne foot, her receding back, and the carried phone clearly legible; use deep focus and strong road-leading geometry to drive the eye toward her path. Shape the night with cool coastal ambience and restrained directional roadside spill, preserving natural detail rather than reducing her to a silhouette.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 수리영이 해안가 도로를 따라 한 발이 허공에 떠 있는 미드 스트라이드(mid-stride) 자세로 달리는 뒷모습.\n\nLOCATION (lock): A coastal bus-stop area with seating facing the open sea, set along the shoreline road. 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): Suri-young continues carrying her mobile phone as she runs away from the bus stop.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPreserve the storyboard’s staging while giving the image a compressed, distant lens feel, visually stacking Suri-young against the coastal roadway and sea-facing bus-stop area. Hold her back and airborne foot at the focal center with shallow depth, allowing a restrained trace of motion in her clothing and grounded leg while the mid-stride pose remains unmistakable. Use soft, low-key night illumination with a narrow rim of coastal ambient light to separate her natural human form from the darker surroundings.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 5 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "스토리보드의 카메라 구도와 버스 정류장의 위치를 잘 재현했으며, 긴 바지를 착용해 캐릭터 레퍼런스의 의상에 더 가깝습니다." }, { "label": "B", "score": 5, "verdict_ko": "미드 스트라이드 자세는 잘 표현되었으나, 캐릭터 레퍼런스와 달리 반바지를 입고 있고 버스 정류장의 배치가 스토리보드와 다소 차이가 납니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S15sh4.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영이 헬멧을 쓰고 있지 않아 캐릭터 레퍼런스와 일치하지 않습니다.", "fix_en": "Add the black bicycle helmet to Suri-young's head, matching the character reference." }, { "issue_ko": "캐릭터 레퍼런스에 있는 크로스백의 검은색 스트랩이 등 부분에 보이지 않습니다.", "fix_en": "Add the black strap of the cross-body bag running diagonally across the back of her jacket." }, { "issue_ko": "버스 정류장 위쪽에 문자가 포함되어 있어 '텍스트 없음(No text)' 규칙을 위반했습니다.", "fix_en": "Remove the illuminated text from the top fascia of the bus stop structure." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the black bicycle helmet to Suri-young's head, matching the character reference.\n- Add the black strap of the cross-body bag running diagonally across the back of her jacket.\n- Remove the illuminated text from the top fascia of the bus stop structure.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S16sh3::variants": { "author_fp": "ae5bc6681dc9a551", "author": { "variants": [ { "approach_ko": "김형사의 시선 위치에서 휴대전화 화면과 혜수의 손을 정면으로 압축한 얕은 심도의 밀착 클로즈업.", "prompt_en": "Photograph from near Detective Kim’s receiving eyeline in an extremely tight, screen-dominant close-up, with Hye-soo’s hand wrapped naturally around the phone and the display plane nearly square to camera. Keep the hand and screen crisply defined while the overnight office dissolves into soft, cool practical-light bokeh, making the offered phone feel immediate and insistent." }, { "approach_ko": "책상 높이의 비스듬한 측면에서 뻗은 손의 동세와 사무실의 깊이를 함께 살린 긴장감 있는 클로즈업.", "prompt_en": "Use a close, desk-height oblique angle that emphasizes the forward extension of Hye-soo’s hand, with the phone seen at a controlled three-quarter angle as it points beyond the camera toward Detective Kim. Let the hand cut diagonally through the frame and retain slightly more office depth behind it, shaped by restrained overhead work light and subtle screen glow for a tense, procedural atmosphere." } ] }, "reused": false }, "S16sh3": { "input_fingerprint": "b9396a3cd1763f8b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 휴대전화 액정 화면을 김형사 쪽으로 내민 혜수의 손 클로즈업.\n\nLOCATION (lock): A police station office with work desks, computer stations, telephones, and an open central workspace suitable for overnight duty. 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): Hye-soo holds her mobile phone with Minsuk’s saved number displayed for Detective Kim.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 휴대전화 액정 화면을 김형사 쪽으로 내민 혜수의 손 클로즈업.\n\nLOCATION (lock): A police station office with work desks, computer stations, telephones, and an open central workspace suitable for overnight duty. 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): Hye-soo holds her mobile phone with Minsuk’s saved number displayed for Detective Kim.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from near Detective Kim’s receiving eyeline in an extremely tight, screen-dominant close-up, with Hye-soo’s hand wrapped naturally around the phone and the display plane nearly square to camera. Keep the hand and screen crisply defined while the overnight office dissolves into soft, cool practical-light bokeh, making the offered phone feel immediate and insistent.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 휴대전화 액정 화면을 김형사 쪽으로 내민 혜수의 손 클로즈업.\n\nLOCATION (lock): A police station office with work desks, computer stations, telephones, and an open central workspace suitable for overnight duty. 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): Hye-soo holds her mobile phone with Minsuk’s saved number displayed for Detective Kim.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close, desk-height oblique angle that emphasizes the forward extension of Hye-soo’s hand, with the phone seen at a controlled three-quarter angle as it points beyond the camera toward Detective Kim. Let the hand cut diagonally through the frame and retain slightly more office depth behind it, shaped by restrained overhead work light and subtle screen glow for a tense, procedural atmosphere.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 휴대전화 액정 화면을 김형사 쪽으로 내민 혜수의 손 클로즈업.\n\nLOCATION (lock): A police station office with work desks, computer stations, telephones, and an open central workspace suitable for overnight duty. 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): Hye-soo holds her mobile phone with Minsuk’s saved number displayed for Detective Kim.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from near Detective Kim’s receiving eyeline in an extremely tight, screen-dominant close-up, with Hye-soo’s hand wrapped naturally around the phone and the display plane nearly square to camera. Keep the hand and screen crisply defined while the overnight office dissolves into soft, cool practical-light bokeh, making the offered phone feel immediate and insistent.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 휴대전화 액정 화면을 김형사 쪽으로 내민 혜수의 손 클로즈업.\n\nLOCATION (lock): A police station office with work desks, computer stations, telephones, and an open central workspace suitable for overnight duty. 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): Hye-soo holds her mobile phone with Minsuk’s saved number displayed for Detective Kim.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close, desk-height oblique angle that emphasizes the forward extension of Hye-soo’s hand, with the phone seen at a controlled three-quarter angle as it points beyond the camera toward Detective Kim. Let the hand cut diagonally through the frame and retain slightly more office depth behind it, shaped by restrained overhead work light and subtle screen glow for a tense, procedural atmosphere.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B02.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S16sh3.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B02.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S16sh3.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B02.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B02.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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": "C", "ranking": [ "C", "D", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "김형사 시점의 완벽한 샷 구도이며, 스마트폰 화면의 UI와 텍스트 레퍼런스를 정확히 재현했습니다." }, { "label": "D", "score": 5, "verdict_ko": "기기를 내미는 방향은 맞으나 혜수의 복장이 다르고 화면 디테일이 훼손되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "화면이 상대가 아닌 본인(카메라) 쪽을 향해 행동이 어색하며 텍스트 디테일이 누락되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "화면이 카메라를 향해 우측 인물에게 내미는 행동이 성립하지 않으며 디테일도 일치하지 않습니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "D", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "김형사 시점의 완벽한 샷 구도이며, 스마트폰 화면의 UI와 텍스트 레퍼런스를 정확히 재현했습니다." }, { "label": "D", "score": 5, "verdict_ko": "기기를 내미는 방향은 맞으나 혜수의 복장이 다르고 화면 디테일이 훼손되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "화면이 상대가 아닌 본인(카메라) 쪽을 향해 행동이 어색하며 텍스트 디테일이 누락되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "화면이 카메라를 향해 우측 인물에게 내미는 행동이 성립하지 않으며 디테일도 일치하지 않습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "C", "B", "A" ], "verdicts": [ { "label": "D", "score": 8, "verdict_ko": "김형사의 어깨를 걸고 화면을 내미는 손 클로즈업 구도를 해부학적 오류 없이 정확하게 연출했습니다." }, { "label": "C", "score": 5, "verdict_ko": "손의 해부학적 구조는 정상이나, 화면이 본인을 향하고 있어 상대에게 내미는 동작 지문을 위반했습니다." }, { "label": "B", "score": 3, "verdict_ko": "인물의 오른팔에 엄지손가락이 왼쪽에 있는 왼손이 결합된 치명적인 해부학적 오류(하드 위반)가 있습니다." }, { "label": "A", "score": 2, "verdict_ko": "손의 구조가 심하게 뭉개진 해부학적 오류(하드 위반)가 있으며, 화면의 텍스트도 알아볼 수 없습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "김형사의 어깨를 걸고 화면을 내미는 손 클로즈업 구도를 해부학적 오류 없이 정확하게 연출했습니다." }, { "label": "B", "score": 5, "verdict_ko": "손의 해부학적 구조는 정상이나, 화면이 본인을 향하고 있어 상대에게 내미는 동작 지문을 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "인물의 오른팔에 엄지손가락이 왼쪽에 있는 왼손이 결합된 치명적인 해부학적 오류(하드 위반)가 있습니다." }, { "label": "D", "score": 2, "verdict_ko": "손의 구조가 심하게 뭉개진 해부학적 오류(하드 위반)가 있으며, 화면의 텍스트도 알아볼 수 없습니다." } ] }, "combined": { "totals": { "A": 11, "B": 9, "C": 10, "D": 7 }, "ranking": [ "A", "C", "B", "D" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 9, "C": 10, "D": 7 }, "selected": "A", "ranking": [ "A", "C", "B", "D" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "김형사 시점의 완벽한 샷 구도이며, 스마트폰 화면의 UI와 텍스트 레퍼런스를 정확히 재현했습니다." }, { "label": "D", "score": 5, "verdict_ko": "기기를 내미는 방향은 맞으나 혜수의 복장이 다르고 화면 디테일이 훼손되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "화면이 상대가 아닌 본인(카메라) 쪽을 향해 행동이 어색하며 텍스트 디테일이 누락되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "화면이 카메라를 향해 우측 인물에게 내미는 행동이 성립하지 않으며 디테일도 일치하지 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B02.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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": "혜수 외의 다른 인물이 등장하는 것을 엄격히 금지한 프롬프트 지시(never anyone else)와 달리, 화면 오른쪽에 다른 인물의 어깨와 뒷모습이 포함되어 있습니다.", "fix_en": "Remove the person on the right side of the frame entirely so that only the arm holding the phone remains." }, { "issue_ko": "휴대전화를 들고 있는 팔의 소매가 캐릭터 레퍼런스에 지정된 혜수의 짙은 남색 셔츠/재킷이 아니라, 배경 레퍼런스 사진에 등장하는 파란색 경찰 근무복 소매로 잘못 렌더링되었습니다.", "fix_en": "Change the visible sleeve on the left arm to match the dark navy fabric and style of Hye-soo's clothing from the character reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the person on the right side of the frame entirely so that only the arm holding the phone remains.\n- Change the visible sleeve on the left arm to match the dark navy fabric and style of Hye-soo's clothing from the character reference.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·A=변형0)", "plate_select": { "candidates": { "A": "L12B01", "B": "L12B02" }, "assigned": "L12B02", "choice": "Candidate B", "confident": false, "reason_ko": "휴대전화를 내미는 손의 클로즈업 샷으로, 배경의 특정 서브 공간을 명확하게 구분할 단서가 텍스트에 없으므로 현재 할당된 후보를 유지합니다.", "kept": "L12B02" } }, "S17sh2::variants": { "author_fp": "d76bf8634b548e96", "author": { "variants": [ { "approach_ko": "담벼락 높이의 낮은 측면 시점에서 고양이를 전경에 크게 두고 불 켜진 창문을 깊은 배경의 시선 종착점으로 잡는다.", "prompt_en": "Photograph from boundary-wall height in a close lateral view, holding the cat in crisp profile as the dominant foreground figure while the illuminated window sits deeper along its exact sightline. Use shallow, selective focus with a gentle falloff into the rooftop surroundings; let the cool dawn ambience contour the dark fur while the interior light provides a restrained warm focal pull." }, { "approach_ko": "약간 높은 원거리 와이드 숏으로 옥상 공간과 담벼락, 작은 고양이, 빛나는 창문의 관계를 정적인 기하학으로 보여준다.", "prompt_en": "Use a restrained wide establishing view from a slightly elevated distance, preserving the rooftop architecture as a strong geometric field. Keep the cat relatively small but clearly readable atop the wall, and compose the illuminated window as the counterweight across the frame, with layered depth and broad focus emphasizing the silent distance between watcher and destination. Balance the cool dawn exterior against the contained interior glow without theatrical contrast." } ] }, "reused": false }, "S17sh2": { "input_fingerprint": "bc731b5e4a92fe48", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn, rooftop room lit from within.\n\nSHOT TEXT (authoritative, Korean): 담벼락 위에 선 검은 고양이가 불 켜진 창문 쪽을 응시하는 정지 상태.\n\nLOCATION (lock): An old villa rooftop and its boundary wall, with a small rooftop dwelling visibly lit from within. 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 rooftop room’s light remains on through dawn while the black cat watches it from the wall.\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn, rooftop room lit from within.\n\nSHOT TEXT (authoritative, Korean): 담벼락 위에 선 검은 고양이가 불 켜진 창문 쪽을 응시하는 정지 상태.\n\nLOCATION (lock): An old villa rooftop and its boundary wall, with a small rooftop dwelling visibly lit from within. 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 rooftop room’s light remains on through dawn while the black cat watches it from the wall.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from boundary-wall height in a close lateral view, holding the cat in crisp profile as the dominant foreground figure while the illuminated window sits deeper along its exact sightline. Use shallow, selective focus with a gentle falloff into the rooftop surroundings; let the cool dawn ambience contour the dark fur while the interior light provides a restrained warm focal pull.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn, rooftop room lit from within.\n\nSHOT TEXT (authoritative, Korean): 담벼락 위에 선 검은 고양이가 불 켜진 창문 쪽을 응시하는 정지 상태.\n\nLOCATION (lock): An old villa rooftop and its boundary wall, with a small rooftop dwelling visibly lit from within. 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 rooftop room’s light remains on through dawn while the black cat watches it from the wall.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained wide establishing view from a slightly elevated distance, preserving the rooftop architecture as a strong geometric field. Keep the cat relatively small but clearly readable atop the wall, and compose the illuminated window as the counterweight across the frame, with layered depth and broad focus emphasizing the silent distance between watcher and destination. Balance the cool dawn exterior against the contained interior glow without theatrical contrast.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 7, "A": 5 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "창문을 응시하는 고양이의 자세가 자연스러우며, 레퍼런스의 초록색 패널 문 디테일을 정확히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "고양이의 시선 방향은 맞으나 자세가 다소 어색하며, 지정된 구조물의 초록색 문 패널 디테일이 누락되었습니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L03B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_villa_rooftop_unit_sel.png" } ], "critique": { "issues": [ { "issue_ko": "옥탑방 외벽이 붉은 벽돌로 되어 있으나, 지정된 구조 레퍼런스 사진에 따라 회색 콘크리트 재질이어야 합니다.", "fix_en": "Change the exterior walls of the rooftop room from red brick to grey concrete to match the structure reference photograph." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the exterior walls of the rooftop room from red brick to grey concrete to match the structure reference photograph.\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.", "ref_mode": "플레이트+seed만 (배경 전용)", "lane_policy": "ab_select_bypass:bg_only", "plate_select": { "candidates": { "A": "L03B01", "B": "L03B02", "C": "L03B03" }, "assigned": "L03B01", "choice": "Candidate A", "confident": true, "reason_ko": "지문에서 요구하는 '담벼락 위에 선 검은 고양이'와 '불 켜진 창문'이 모두 정확히 묘사되어 있으며, 이를 충족하는 현재 할당본을 유지하는 것이 적합합니다.", "kept": "L03B01" } }, "S18sh8::variants": { "author_fp": "05795143a95dcfed", "author": { "variants": [ { "approach_ko": "거울을 정면으로 밀착해 붉은 원을 중심에 고립시키는 불안한 주관 시점.", "prompt_en": "Frame the mirror in a tight, nearly frontal point-of-view close shot, with the reflected bathroom wall occupying almost the entire image. Hold the vivid red circle in crisp focus near the compositional center, using restrained, cool early-morning ambient light and shallow surrounding falloff to make the mark feel immediate and unnervingly isolated." }, { "approach_ko": "거울 가장자리와 욕실의 실제 공간을 함께 담아 반사 속 붉은 원의 모순을 강조하는 비스듬한 주관 시점.", "prompt_en": "Photograph from an oblique point-of-view angle, keeping a clear edge of the mirror in the foreground while the reflected wall recedes into deeper space. Use layered, deep-focus composition and soft directional morning illumination so the vivid red circle reads sharply within the reflection, while the mirror boundary creates quiet spatial tension." } ] }, "reused": false }, "S18sh8": { "input_fingerprint": "6fea4a0ca05461ac", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 거울에 비친 욕실 벽면에 선명한 붉은 원이 그려져 있는 시점 쇼트.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young now has her doll-adorned backpack with her. A vivid red circle appears on the bathroom wall in the mirror, but it vanishes when she turns to look directly at 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 거울에 비친 욕실 벽면에 선명한 붉은 원이 그려져 있는 시점 쇼트.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young now has her doll-adorned backpack with her. A vivid red circle appears on the bathroom wall in the mirror, but it vanishes when she turns to look directly at 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame the mirror in a tight, nearly frontal point-of-view close shot, with the reflected bathroom wall occupying almost the entire image. Hold the vivid red circle in crisp focus near the compositional center, using restrained, cool early-morning ambient light and shallow surrounding falloff to make the mark feel immediate and unnervingly isolated.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 거울에 비친 욕실 벽면에 선명한 붉은 원이 그려져 있는 시점 쇼트.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young now has her doll-adorned backpack with her. A vivid red circle appears on the bathroom wall in the mirror, but it vanishes when she turns to look directly at 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from an oblique point-of-view angle, keeping a clear edge of the mirror in the foreground while the reflected wall recedes into deeper space. Use layered, deep-focus composition and soft directional morning illumination so the vivid red circle reads sharply within the reflection, while the mirror boundary creates quiet spatial tension.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "거울에 비친 욕실 벽면과 선명한 붉은 원을 지시문대로 정확히 연출했으나 반사된 공간 구조가 레퍼런스와 약간 다릅니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스의 복도 구조를 반사한 점은 좋으나, 왼쪽 벽에 원본에 없는 두 번째 거울과 창문이 생성되어 공간적 왜곡이 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B05.png" } ], "critique": { "issues": [] }, "fix_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B05", "choice": "Candidate E", "confident": true, "reason_ko": "지문에서 '욕실'과 '붉은 원'을 명시하고 있으며, 해당 후보 이미지는 욕실 공간을 보여줄 뿐만 아니라 이미지 상에 붉은 원이 표시되어 있어 지문과 가장 완벽하게 일치합니다.", "kept": "L04B05" } }, "S18sh11::variants": { "author_fp": "876c8bc7bdecac6a", "author": { "variants": [ { "approach_ko": "사진 면을 정면으로 압박하는 극단적 클로즈업으로, 환한 표정의 기괴한 붕괴를 피할 곳 없이 제시한다.", "prompt_en": "Photograph the image nearly straight-on in an oppressive extreme close-up, with the face dominating the frame and the surrounding photograph reduced to narrow contextual edges. Keep the print plane crisp and materially convincing, rendering the momentary downward smearing and compression with unsettling clarity under soft, even early-morning illumination." }, { "approach_ko": "사진 표면을 비스듬히 훑는 얕은 초점의 클로즈업으로, 일상적인 인화지 위의 순간적 왜곡을 불안하게 포착한다.", "prompt_en": "Use a close, sharply oblique view that looks across the photograph’s surface, creating a receding plane and shallow focus that locks onto the distorted smiling face while the nearer and farther portions fall gently soft. Let restrained, raking early-morning light reveal the print’s physical surface, making the grotesque flowing distortion feel fleeting, intimate, and captured in-camera rather than graphically exaggerated." } ] }, "reused": false }, "S18sh11": { "input_fingerprint": "588345b881a27175", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 사진 속 강민숙의 환하게 웃는 얼굴이 기괴하게 흘러내린 채 뭉개져 있는 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young continues carrying the doll-adorned backpack. Minsuk’s face appears distorted and melting only momentarily; the photograph is normal again when Suri-young looks once more.\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 사진 속 강민숙의 환하게 웃는 얼굴이 기괴하게 흘러내린 채 뭉개져 있는 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young continues carrying the doll-adorned backpack. Minsuk’s face appears distorted and melting only momentarily; the photograph is normal again when Suri-young looks once more.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the image nearly straight-on in an oppressive extreme close-up, with the face dominating the frame and the surrounding photograph reduced to narrow contextual edges. Keep the print plane crisp and materially convincing, rendering the momentary downward smearing and compression with unsettling clarity under soft, even early-morning illumination.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 사진 속 강민숙의 환하게 웃는 얼굴이 기괴하게 흘러내린 채 뭉개져 있는 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young continues carrying the doll-adorned backpack. Minsuk’s face appears distorted and melting only momentarily; the photograph is normal again when Suri-young looks once more.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close, sharply oblique view that looks across the photograph’s surface, creating a receding plane and shallow focus that locks onto the distorted smiling face while the nearer and farther portions fall gently soft. Let restrained, raking early-morning light reveal the print’s physical surface, making the grotesque flowing distortion feel fleeting, intimate, and captured in-camera rather than graphically exaggerated.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 7, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시문대로 사진 속 얼굴이 기괴하게 흘러내리는 왜곡된 모습을 성공적으로 연출했으며, 배경의 장소 일치도도 뛰어납니다." }, { "label": "A", "score": 3, "verdict_ko": "클로즈업 구도는 좋으나, 얼굴이 흘러내리는 묘사가 누락된 채 단순히 구겨진 비닐 반사처럼 표현되어 핵심 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B05.png" } ], "critique": { "issues": [] }, "fix_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B05", "choice": "Candidate E", "confident": true, "reason_ko": "원문에서 '사진 속 강민숙의 환하게 웃는 얼굴'이라는 묘사가 등장하므로 거실 벽면에 가족 사진(액자)이 걸려 있는 유일한 후보인 Candidate E가 적합합니다.", "kept": "L04B05" } }, "S18sh12::variants": { "author_fp": "fcad8d065827a1d7", "author": { "variants": [ { "approach_ko": "뺨 높이의 측면 초근접 촬영으로 손바닥의 충격과 밀려나는 피부를 얕은 심도로 날카롭게 포착한다.", "prompt_en": "Photograph the impact in an extreme side close-up at cheek height, with the palm, compressed cheek, and displaced skin forming a sharp diagonal across the frame. Use a close-focus, shallow-depth lens feel: the contact plane is crisply frozen while the far contours of the face and hand fall gently out of focus. Soft lateral early-morning light grazes the skin, revealing the physical compression without glamorizing it." }, { "approach_ko": "정면에 가까운 타이트 클로즈업과 깊은 초점으로 얼굴의 반응과 자기 손의 타격 동작을 동시에 강조한다.", "prompt_en": "Frame a tight, nearly frontal close-up with the face held centrally and the striking hand entering forcefully from the side, preserving enough surrounding facial structure to read the involuntary reaction at the exact instant of contact. Favor a compressed, observational lens feel and moderately deep focus so the eyes, palm, and distorted cheek remain legible together. Use restrained interior illumination with subtle early-morning ambient lift, keeping the image stark, immediate, and psychologically confrontational." } ] }, "reused": false }, "S18sh12": { "input_fingerprint": "140f6790b4d88f07", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손바닥이 자신의 뺨을 강하게 가격하여 피부가 밀려난 mid-impact 찰나의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young still has the doll-adorned backpack with her; the family photograph has returned to its normal 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손바닥이 자신의 뺨을 강하게 가격하여 피부가 밀려난 mid-impact 찰나의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young still has the doll-adorned backpack with her; the family photograph has returned to its normal 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the impact in an extreme side close-up at cheek height, with the palm, compressed cheek, and displaced skin forming a sharp diagonal across the frame. Use a close-focus, shallow-depth lens feel: the contact plane is crisply frozen while the far contours of the face and hand fall gently out of focus. Soft lateral early-morning light grazes the skin, revealing the physical compression without glamorizing it.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손바닥이 자신의 뺨을 강하게 가격하여 피부가 밀려난 mid-impact 찰나의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young still has the doll-adorned backpack with her; the family photograph has returned to its normal 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight, nearly frontal close-up with the face held centrally and the striking hand entering forcefully from the side, preserving enough surrounding facial structure to read the involuntary reaction at the exact instant of contact. Favor a compressed, observational lens feel and moderately deep focus so the eyes, palm, and distorted cheek remain legible together. Use restrained interior illumination with subtle early-morning ambient lift, keeping the image stark, immediate, and psychologically confrontational.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손바닥이 자신의 뺨을 강하게 가격하여 피부가 밀려난 mid-impact 찰나의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young still has the doll-adorned backpack with her; the family photograph has returned to its normal 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the impact in an extreme side close-up at cheek height, with the palm, compressed cheek, and displaced skin forming a sharp diagonal across the frame. Use a close-focus, shallow-depth lens feel: the contact plane is crisply frozen while the far contours of the face and hand fall gently out of focus. Soft lateral early-morning light grazes the skin, revealing the physical compression without glamorizing it.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손바닥이 자신의 뺨을 강하게 가격하여 피부가 밀려난 mid-impact 찰나의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a main bedroom, a separate curtained bedroom, a small bathroom with a mirror, and a living-dining area displaying a family photograph. 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): Suri-young still has the doll-adorned backpack with her; the family photograph has returned to its normal 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight, nearly frontal close-up with the face held centrally and the striking hand entering forcefully from the side, preserving enough surrounding facial structure to read the involuntary reaction at the exact instant of contact. Favor a compressed, observational lens feel and moderately deep focus so the eyes, palm, and distorted cheek remain legible together. Use restrained interior illumination with subtle early-morning ambient lift, keeping the image stark, immediate, and psychologically confrontational.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B05.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S18sh12.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B05.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S18sh12.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B05.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B05.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "C", "D", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "자신의 뺨을 강하게 가격하여 피부가 밀리는 찰나의 클로즈업을 정확히 연출했으며, 인물과 배경, 인형 배낭 소품까지 모두 일치합니다." }, { "label": "C", "score": 4, "verdict_ko": "클로즈업과 배경 조건은 충족했으나, 손이 뺨에 가볍게 닿아 있을 뿐 강하게 가격하는 타격의 순간이 표현되지 않았습니다." }, { "label": "D", "score": 3, "verdict_ko": "양손으로 얼굴을 감싸고 있는 포즈를 취해 '손바닥이 뺨을 강하게 가격하는' 핵심 액션 지시를 따르지 않았습니다." }, { "label": "A", "score": 2, "verdict_ko": "'자신의 뺨'을 때려야 한다는 지시를 어기고 제3자가 등장하는 치명적인 규정 위반이 있습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "C", "D", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "자신의 뺨을 강하게 가격하여 피부가 밀리는 찰나의 클로즈업을 정확히 연출했으며, 인물과 배경, 인형 배낭 소품까지 모두 일치합니다." }, { "label": "C", "score": 4, "verdict_ko": "클로즈업과 배경 조건은 충족했으나, 손이 뺨에 가볍게 닿아 있을 뿐 강하게 가격하는 타격의 순간이 표현되지 않았습니다." }, { "label": "D", "score": 3, "verdict_ko": "양손으로 얼굴을 감싸고 있는 포즈를 취해 '손바닥이 뺨을 강하게 가격하는' 핵심 액션 지시를 따르지 않았습니다." }, { "label": "A", "score": 2, "verdict_ko": "'자신의 뺨'을 때려야 한다는 지시를 어기고 제3자가 등장하는 치명적인 규정 위반이 있습니다." } ] }, "reverse_raw": { "winner": "C", "ranking": [ "C", "B", "A", "D" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "강하게 가격하는 찰나가 아니라 양손으로 뺨을 가볍게 감싸고 있는 정적인 포즈로 연출되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "타격으로 피부가 밀려난 순간은 좋으나, 손의 각도와 구도상 타인의 손처럼 보일 여지가 다소 있습니다." }, { "label": "C", "score": 8, "verdict_ko": "자신의 뺨을 강하게 가격하여 피부가 밀려나는 역동적인 클로즈업 묘사가 가장 정확하고 자연스럽게 구현되었습니다." }, { "label": "D", "score": 3, "verdict_ko": "프롬프트에 명시되지 않은 타인이 등장하여 주인공을 때리는 심각한 규정 위반이 발생했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "C", "D", "A" ], "verdicts": [ { "label": "D", "score": 5, "verdict_ko": "강하게 가격하는 찰나가 아니라 양손으로 뺨을 가볍게 감싸고 있는 정적인 포즈로 연출되었습니다." }, { "label": "C", "score": 7, "verdict_ko": "타격으로 피부가 밀려난 순간은 좋으나, 손의 각도와 구도상 타인의 손처럼 보일 여지가 다소 있습니다." }, { "label": "B", "score": 8, "verdict_ko": "자신의 뺨을 강하게 가격하여 피부가 밀려나는 역동적인 클로즈업 묘사가 가장 정확하고 자연스럽게 구현되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "프롬프트에 명시되지 않은 타인이 등장하여 주인공을 때리는 심각한 규정 위반이 발생했습니다." } ] }, "combined": { "totals": { "A": 5, "B": 15, "C": 11, "D": 8 }, "ranking": [ "B", "C", "D", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 5, "B": 15, "C": 11, "D": 8 }, "selected": "B", "ranking": [ "B", "C", "D", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "자신의 뺨을 강하게 가격하여 피부가 밀리는 찰나의 클로즈업을 정확히 연출했으며, 인물과 배경, 인형 배낭 소품까지 모두 일치합니다." }, { "label": "C", "score": 4, "verdict_ko": "클로즈업과 배경 조건은 충족했으나, 손이 뺨에 가볍게 닿아 있을 뿐 강하게 가격하는 타격의 순간이 표현되지 않았습니다." }, { "label": "D", "score": 3, "verdict_ko": "양손으로 얼굴을 감싸고 있는 포즈를 취해 '손바닥이 뺨을 강하게 가격하는' 핵심 액션 지시를 따르지 않았습니다." }, { "label": "A", "score": 2, "verdict_ko": "'자신의 뺨'을 때려야 한다는 지시를 어기고 제3자가 등장하는 치명적인 규정 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B05.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프레임 우측에서 들어와 뺨에 닿아 있는 손은 해부학적으로 오른손의 형태를 띠고 있어 타인의 손처럼 보이며, 수리영이 '자신의 뺨'을 때린다는 지시사항을 위반합니다.", "fix_en": "Replace the hand on the cheek with Suri-young's own left hand captured in a forceful, mid-impact slapping motion." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the hand on the cheek with Suri-young's own left hand captured in a forceful, mid-impact slapping motion.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·B=변형1)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B05", "choice": "Candidate E", "confident": false, "reason_ko": "샷 텍스트가 특정 공간(예: 화장실, 거실 등)을 명시하지 않아 배경을 확정할 단서가 부족하므로, 규칙에 따라 현재 할당된 공간을 유지합니다.", "kept": "L04B05" } }, "S19sh2::variants": { "author_fp": "5fd1c10eff262758", "author": { "variants": [ { "approach_ko": "수리영의 눈높이에서 정면에 가깝게 밀착해 그녀의 가로막는 자세와 올려다보는 시선을 중심에 둔 상체 투숏.", "prompt_en": "Frame a tight upper-body two-shot from near Suri-young’s eye level, looking almost frontally toward her as she occupies the visual center and blocks the employee’s path. Keep the employee close behind and above her in the composition, preserving their height relationship while making her upward gaze and resolute posture the dramatic focus. Use intimate depth separation and soft early-morning supermarket illumination, with the interior receding unobtrusively behind them." }, { "approach_ko": "두 사람을 옆에서 포착해 통로를 막는 수리영의 몸과 위아래로 맞물린 시선 관계를 선명한 측면 구도로 강조.", "prompt_en": "Photograph the encounter as a lateral upper-body profile two-shot, placing Suri-young and the employee on opposing sides of the frame so her body visibly interrupts his forward line. Use a slightly lower camera height and a restrained, compressed lens feel to emphasize her lifted chin and the vertical distance between their faces. Hold both figures clearly within the same depth plane under directional early-morning ambient light, letting the supermarket geometry create quiet tension around their standoff." } ] }, "reused": false }, "S19sh2": { "input_fingerprint": "cf7fdda7a5135d97", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 안, 수리영이 마트 남자 직원 앞을 가로막고 선 채 올려다보는 상체 구도.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young continues carrying her doll-adorned backpack while questioning the mart employee.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 마트 남자 직원 (한국인 성인 남성, 짙은색 머리, 평범한 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 안, 수리영이 마트 남자 직원 앞을 가로막고 선 채 올려다보는 상체 구도.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young continues carrying her doll-adorned backpack while questioning the mart employee.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 마트 남자 직원 (한국인 성인 남성, 짙은색 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight upper-body two-shot from near Suri-young’s eye level, looking almost frontally toward her as she occupies the visual center and blocks the employee’s path. Keep the employee close behind and above her in the composition, preserving their height relationship while making her upward gaze and resolute posture the dramatic focus. Use intimate depth separation and soft early-morning supermarket illumination, with the interior receding unobtrusively behind them.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 안, 수리영이 마트 남자 직원 앞을 가로막고 선 채 올려다보는 상체 구도.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young continues carrying her doll-adorned backpack while questioning the mart employee.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 마트 남자 직원 (한국인 성인 남성, 짙은색 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the encounter as a lateral upper-body profile two-shot, placing Suri-young and the employee on opposing sides of the frame so her body visibly interrupts his forward line. Use a slightly lower camera height and a restrained, compressed lens feel to emphasize her lifted chin and the vertical distance between their faces. Hold both figures clearly within the same depth plane under directional early-morning ambient light, letting the supermarket geometry create quiet tension around their standoff.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 안, 수리영이 마트 남자 직원 앞을 가로막고 선 채 올려다보는 상체 구도.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young continues carrying her doll-adorned backpack while questioning the mart employee.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 마트 남자 직원 (한국인 성인 남성, 짙은색 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight upper-body two-shot from near Suri-young’s eye level, looking almost frontally toward her as she occupies the visual center and blocks the employee’s path. Keep the employee close behind and above her in the composition, preserving their height relationship while making her upward gaze and resolute posture the dramatic focus. Use intimate depth separation and soft early-morning supermarket illumination, with the interior receding unobtrusively behind them.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 안, 수리영이 마트 남자 직원 앞을 가로막고 선 채 올려다보는 상체 구도.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young continues carrying her doll-adorned backpack while questioning the mart employee.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 마트 남자 직원 (한국인 성인 남성, 짙은색 머리, 평범한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the encounter as a lateral upper-body profile two-shot, placing Suri-young and the employee on opposing sides of the frame so her body visibly interrupts his forward line. Use a slightly lower camera height and a restrained, compressed lens feel to emphasize her lifted chin and the vertical distance between their faces. Hold both figures clearly within the same depth plane under directional early-morning ambient light, letting the supermarket geometry create quiet tension around their standoff.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B04.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S19sh2.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B04.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S19sh2.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B04.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B04.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '상체 구도'를 정확히 구현했으며, 올려다보는 시선과 인형 달린 가방 등 핵심 요소를 가장 충실히 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "캐릭터와 소품 디테일은 우수하나, 프레이밍이 지시된 '상체 구도'보다 넓게 잡혀 우선순위에서 밀립니다." }, { "label": "C", "score": 3, "verdict_ko": "프레이밍이 넓을 뿐만 아니라, 실외에 있어야 할 자전거 보관소가 마트 내부에 배치되는 치명적인 공간적 오류가 있습니다." }, { "label": "D", "score": 5, "verdict_ko": "인물간의 배치는 무난하지만 프레이밍이 '상체 구도'를 크게 벗어나 지나치게 넓게 촬영되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '상체 구도'를 정확히 구현했으며, 올려다보는 시선과 인형 달린 가방 등 핵심 요소를 가장 충실히 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "캐릭터와 소품 디테일은 우수하나, 프레이밍이 지시된 '상체 구도'보다 넓게 잡혀 우선순위에서 밀립니다." }, { "label": "C", "score": 3, "verdict_ko": "프레이밍이 넓을 뿐만 아니라, 실외에 있어야 할 자전거 보관소가 마트 내부에 배치되는 치명적인 공간적 오류가 있습니다." }, { "label": "D", "score": 5, "verdict_ko": "인물간의 배치는 무난하지만 프레이밍이 '상체 구도'를 크게 벗어나 지나치게 넓게 촬영되었습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "의상은 정확히 재현했으나 무릎까지 노출되는 넓은 화각으로 인해 상체 구도(Priority 2) 지시를 크게 위반함." }, { "label": "B", "score": 4, "verdict_ko": "상체 구도보다 넓은 화각(Priority 2)을 보이며, 수리영의 파란색 재킷 의상(Priority 4)도 일치하지 않음." }, { "label": "C", "score": 6, "verdict_ko": "의상과 소품 재현율은 훌륭하나 허벅지까지 노출되어 엄격한 상체 구도(Priority 2) 기준에서 D에게 밀림." }, { "label": "D", "score": 7, "verdict_ko": "파란색 재킷 의상(Priority 4) 오류가 있으나, 유일하게 지시된 상체 구도(Priority 2)를 정확히 구현하여 우승함." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "D", "score": 5, "verdict_ko": "의상은 정확히 재현했으나 무릎까지 노출되는 넓은 화각으로 인해 상체 구도(Priority 2) 지시를 크게 위반함." }, { "label": "C", "score": 4, "verdict_ko": "상체 구도보다 넓은 화각(Priority 2)을 보이며, 수리영의 파란색 재킷 의상(Priority 4)도 일치하지 않음." }, { "label": "B", "score": 6, "verdict_ko": "의상과 소품 재현율은 훌륭하나 허벅지까지 노출되어 엄격한 상체 구도(Priority 2) 기준에서 D에게 밀림." }, { "label": "A", "score": 7, "verdict_ko": "파란색 재킷 의상(Priority 4) 오류가 있으나, 유일하게 지시된 상체 구도(Priority 2)를 정확히 구현하여 우승함." } ] }, "combined": { "totals": { "A": 14, "B": 12, "C": 7, "D": 10 }, "ranking": [ "A", "B", "D", "C" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 12, "C": 7, "D": 10 }, "selected": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '상체 구도'를 정확히 구현했으며, 올려다보는 시선과 인형 달린 가방 등 핵심 요소를 가장 충실히 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "캐릭터와 소품 디테일은 우수하나, 프레이밍이 지시된 '상체 구도'보다 넓게 잡혀 우선순위에서 밀립니다." }, { "label": "C", "score": 3, "verdict_ko": "프레이밍이 넓을 뿐만 아니라, 실외에 있어야 할 자전거 보관소가 마트 내부에 배치되는 치명적인 공간적 오류가 있습니다." }, { "label": "D", "score": 5, "verdict_ko": "인물간의 배치는 무난하지만 프레이밍이 '상체 구도'를 크게 벗어나 지나치게 넓게 촬영되었습니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B04.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "마트 천장의 구역 안내판에 글자가 포함되어 있어 'no text' 규칙을 위반했습니다.", "fix_en": "Remove all text from the overhead aisle signs." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text from the overhead aisle signs.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·A=변형0)", "plate_select": { "candidates": { "A": "L08B01", "B": "L08B02", "C": "L08B03", "D": "L08B04" }, "assigned": "L08B04", "choice": "Candidate D", "confident": true, "reason_ko": "\"마트 안\"이라는 지문에 따라 마트 내부 공간이 필요함.", "kept": "L08B04" } }, "S19sh4::variants": { "author_fp": "b410f61d40f7c080", "author": { "variants": [ { "approach_ko": "처마 아래의 낮은 근접 앵글로 수리영의 올려다보는 얼굴과 손가락, CCTV를 강한 대각선으로 연결한다.", "prompt_en": "Photograph from a low, close position beneath the eave, using a moderately wide lens feel so her upturned face, pointing hand, and the surveillance camera form a strong diagonal through the frame. Keep her mobile phone and doll-adorned backpack naturally legible without distracting from the held gesture. Let soft early-morning light skim across her face and hand while the underside of the eave retains quiet, dimensional shadow; render the supermarket exterior with crisp live-action texture and restrained depth." }, { "approach_ko": "한발 물러선 관찰자 시점으로 수리영을 외벽과 CCTV 배열 속에 작게 두어 멈춰 선 순간과 감시 공간을 강조한다.", "prompt_en": "Use a distant, eye-level observational composition that places her full figure within the geometry of the supermarket exterior, with the surveillance-covered wall and eave dominating the frame and the bicycle parking contributing layered depth. Preserve clear separation between her raised pointing arm and the camera she indicates, while her phone and doll-adorned backpack remain visible as carried objects. Favor compressed, orderly planes and cool, even early-morning illumination, creating a quiet sense of being watched rather than an intimate portrait." } ] }, "reused": false }, "S19sh4": { "input_fingerprint": "849af667e5b988f0", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 CCTV가 설치된 처마 아래, 수리영이 고개를 든 채 CCTV 카메라를 손가락으로 가리키고 멈춰 서 있는 순간.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young still carries the doll-adorned backpack and her mobile phone as she identifies the exterior CCTV position and calls Hye-soo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 CCTV가 설치된 처마 아래, 수리영이 고개를 든 채 CCTV 카메라를 손가락으로 가리키고 멈춰 서 있는 순간.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young still carries the doll-adorned backpack and her mobile phone as she identifies the exterior CCTV position and calls Hye-soo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a low, close position beneath the eave, using a moderately wide lens feel so her upturned face, pointing hand, and the surveillance camera form a strong diagonal through the frame. Keep her mobile phone and doll-adorned backpack naturally legible without distracting from the held gesture. Let soft early-morning light skim across her face and hand while the underside of the eave retains quiet, dimensional shadow; render the supermarket exterior with crisp live-action texture and restrained depth.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 CCTV가 설치된 처마 아래, 수리영이 고개를 든 채 CCTV 카메라를 손가락으로 가리키고 멈춰 서 있는 순간.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young still carries the doll-adorned backpack and her mobile phone as she identifies the exterior CCTV position and calls Hye-soo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a distant, eye-level observational composition that places her full figure within the geometry of the supermarket exterior, with the surveillance-covered wall and eave dominating the frame and the bicycle parking contributing layered depth. Preserve clear separation between her raised pointing arm and the camera she indicates, while her phone and doll-adorned backpack remain visible as carried objects. Favor compressed, orderly planes and cool, even early-morning illumination, creating a quiet sense of being watched rather than an intimate portrait.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 CCTV가 설치된 처마 아래, 수리영이 고개를 든 채 CCTV 카메라를 손가락으로 가리키고 멈춰 서 있는 순간.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young still carries the doll-adorned backpack and her mobile phone as she identifies the exterior CCTV position and calls Hye-soo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a low, close position beneath the eave, using a moderately wide lens feel so her upturned face, pointing hand, and the surveillance camera form a strong diagonal through the frame. Keep her mobile phone and doll-adorned backpack naturally legible without distracting from the held gesture. Let soft early-morning light skim across her face and hand while the underside of the eave retains quiet, dimensional shadow; render the supermarket exterior with crisp live-action texture and restrained depth.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖 CCTV가 설치된 처마 아래, 수리영이 고개를 든 채 CCTV 카메라를 손가락으로 가리키고 멈춰 서 있는 순간.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with bicycle parking outside, broad retail aisles inside, and an exterior wall covered by surveillance cameras. 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): Suri-young still carries the doll-adorned backpack and her mobile phone as she identifies the exterior CCTV position and calls Hye-soo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a distant, eye-level observational composition that places her full figure within the geometry of the supermarket exterior, with the surveillance-covered wall and eave dominating the frame and the bicycle parking contributing layered depth. Preserve clear separation between her raised pointing arm and the camera she indicates, while her phone and doll-adorned backpack remain visible as carried objects. Favor compressed, orderly planes and cool, even early-morning illumination, creating a quiet sense of being watched rather than an intimate portrait.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B03.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S19sh4.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B03.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S19sh4.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B03.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B03.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 장소와 의상을 정확히 재현했으며, CCTV를 가리키며 휴대폰을 든 요구 동작을 충실히 구현함." }, { "label": "B", "score": 5, "verdict_ko": "장소와 동작은 잘 표현되었으나, 캐릭터의 의상(검은색 재킷)이 레퍼런스와 불일치함." }, { "label": "C", "score": 4, "verdict_ko": "휴대폰과 의상은 일치하나, 레퍼런스의 마트 외관과 구조가 아닌 다른 장소가 배경으로 생성됨." }, { "label": "D", "score": 3, "verdict_ko": "휴대폰을 들고 있지 않으며, 기둥에 레퍼런스에 없는 다수의 CCTV가 임의로 추가되는 오류가 있음." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 장소와 의상을 정확히 재현했으며, CCTV를 가리키며 휴대폰을 든 요구 동작을 충실히 구현함." }, { "label": "B", "score": 5, "verdict_ko": "장소와 동작은 잘 표현되었으나, 캐릭터의 의상(검은색 재킷)이 레퍼런스와 불일치함." }, { "label": "C", "score": 4, "verdict_ko": "휴대폰과 의상은 일치하나, 레퍼런스의 마트 외관과 구조가 아닌 다른 장소가 배경으로 생성됨." }, { "label": "D", "score": 3, "verdict_ko": "휴대폰을 들고 있지 않으며, 기둥에 레퍼런스에 없는 다수의 CCTV가 임의로 추가되는 오류가 있음." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "정확한 장소 묘사와 함께 지정된 의상, 소품(휴대폰, 인형 백팩), 그리고 CCTV를 가리키는 동작을 가장 충실하게 구현했습니다." }, { "label": "C", "score": 5, "verdict_ko": "배경과 동작의 연출은 훌륭하나, 레퍼런스와 일치하지 않는 의상(검은 재킷)과 긴 머리스타일이 감점 요인입니다." }, { "label": "A", "score": 4, "verdict_ko": "장소는 일치하지만 구도가 너무 넓어 피사체가 작고, 휴대폰이 보이지 않으며 백팩의 인형이 사람의 몸처럼 묘사되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "의상과 소품은 양호하게 반영되었으나, 레퍼런스 사진의 마트 건축 구조와 완전히 다른 장소를 묘사하여 위치 조건에 크게 위배됩니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "정확한 장소 묘사와 함께 지정된 의상, 소품(휴대폰, 인형 백팩), 그리고 CCTV를 가리키는 동작을 가장 충실하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "배경과 동작의 연출은 훌륭하나, 레퍼런스와 일치하지 않는 의상(검은 재킷)과 긴 머리스타일이 감점 요인입니다." }, { "label": "D", "score": 4, "verdict_ko": "장소는 일치하지만 구도가 너무 넓어 피사체가 작고, 휴대폰이 보이지 않으며 백팩의 인형이 사람의 몸처럼 묘사되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "의상과 소품은 양호하게 반영되었으나, 레퍼런스 사진의 마트 건축 구조와 완전히 다른 장소를 묘사하여 위치 조건에 크게 위배됩니다." } ] }, "combined": { "totals": { "A": 14, "B": 10, "C": 7, "D": 7 }, "ranking": [ "A", "B", "C", "D" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 10, "C": 7, "D": 7 }, "selected": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 장소와 의상을 정확히 재현했으며, CCTV를 가리키며 휴대폰을 든 요구 동작을 충실히 구현함." }, { "label": "B", "score": 5, "verdict_ko": "장소와 동작은 잘 표현되었으나, 캐릭터의 의상(검은색 재킷)이 레퍼런스와 불일치함." }, { "label": "C", "score": 4, "verdict_ko": "휴대폰과 의상은 일치하나, 레퍼런스의 마트 외관과 구조가 아닌 다른 장소가 배경으로 생성됨." }, { "label": "D", "score": 3, "verdict_ko": "휴대폰을 들고 있지 않으며, 기둥에 레퍼런스에 없는 다수의 CCTV가 임의로 추가되는 오류가 있음." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L08B03.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "캐릭터의 같은 쪽(오른쪽)에 팔이 두 개(가리키는 팔과 휴대폰을 든 팔)가 생성되어 해부학적으로 어색합니다.", "fix_en": "Remove the extra arm holding the phone in the foreground, keeping only one right arm pointing upward, and place the mobile phone in her left hand." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the extra arm holding the phone in the foreground, keeping only one right arm pointing upward, and place the mobile phone in her left hand.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·A=변형0)", "plate_select": { "candidates": { "A": "L08B01", "B": "L08B02", "C": "L08B03", "D": "L08B04" }, "assigned": "L08B03", "choice": "Candidate C", "confident": true, "reason_ko": "마트 밖 처마 아래에 설치된 CCTV 카메라가 화면에 크게 잡혀 있어, 인물이 고개를 들어 카메라를 가리키는 상황을 연출하기에 가장 적합한 서브공간입니다.", "kept": "L08B03" } }, "S20sh2::variants": { "author_fp": "33b50638e453986b", "author": { "variants": [ { "approach_ko": "모니터 바로 옆의 시선 높이에서 정면에 가깝게 밀착해, 미간과 눈빛의 긴장을 얕은 심도로 포착한다.", "prompt_en": "Shoot from beside the monitor at her eye level in an extremely tight, near-frontal facial close-up. Keep her eyes and furrowed brow critically sharp while the office equipment falls into soft blur behind her; let restrained monitor light model her face, with just enough shoulder visible to preserve the sense of the backpack she is carrying." }, { "approach_ko": "약간 높은 측면의 압축된 클로즈업으로 얼굴 윤곽과 모니터를 향한 응시를 강조하고, 장비를 전경 층으로 활용한다.", "prompt_en": "Use a compressed three-quarter profile close-up from slightly above, placing her face off-center with breathing room along the direction of her gaze. Frame her through softly defocused edges of the playback equipment for a tense, enclosed composition, with directional screen illumination tracing her cheek and tightened brow while the backpack strap remains at the lower edge of frame." } ] }, "reused": false }, "S20sh2": { "input_fingerprint": "7289d620aea98985", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 보안직원과 혜수 뒤에 선 수리영이 미간을 찌푸린 채 모니터를 응시하는 얼굴 클로즈업.\n\nLOCATION (lock): A supermarket administrative and security office organized around CCTV monitors, playback equipment, and staff workstations. 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): Suri-young continues carrying her doll-adorned backpack while watching the CCTV playback.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 보안직원과 혜수 뒤에 선 수리영이 미간을 찌푸린 채 모니터를 응시하는 얼굴 클로즈업.\n\nLOCATION (lock): A supermarket administrative and security office organized around CCTV monitors, playback equipment, and staff workstations. 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): Suri-young continues carrying her doll-adorned backpack while watching the CCTV playback.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from beside the monitor at her eye level in an extremely tight, near-frontal facial close-up. Keep her eyes and furrowed brow critically sharp while the office equipment falls into soft blur behind her; let restrained monitor light model her face, with just enough shoulder visible to preserve the sense of the backpack she is carrying.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 보안직원과 혜수 뒤에 선 수리영이 미간을 찌푸린 채 모니터를 응시하는 얼굴 클로즈업.\n\nLOCATION (lock): A supermarket administrative and security office organized around CCTV monitors, playback equipment, and staff workstations. 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): Suri-young continues carrying her doll-adorned backpack while watching the CCTV playback.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed three-quarter profile close-up from slightly above, placing her face off-center with breathing room along the direction of her gaze. Frame her through softly defocused edges of the playback equipment for a tense, enclosed composition, with directional screen illumination tracing her cheek and tightened brow while the backpack strap remains at the lower edge of frame.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 보안직원과 혜수 뒤에 선 수리영이 미간을 찌푸린 채 모니터를 응시하는 얼굴 클로즈업.\n\nLOCATION (lock): A supermarket administrative and security office organized around CCTV monitors, playback equipment, and staff workstations. 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): Suri-young continues carrying her doll-adorned backpack while watching the CCTV playback.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from beside the monitor at her eye level in an extremely tight, near-frontal facial close-up. Keep her eyes and furrowed brow critically sharp while the office equipment falls into soft blur behind her; let restrained monitor light model her face, with just enough shoulder visible to preserve the sense of the backpack she is carrying.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 보안직원과 혜수 뒤에 선 수리영이 미간을 찌푸린 채 모니터를 응시하는 얼굴 클로즈업.\n\nLOCATION (lock): A supermarket administrative and security office organized around CCTV monitors, playback equipment, and staff workstations. 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): Suri-young continues carrying her doll-adorned backpack while watching the CCTV playback.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed three-quarter profile close-up from slightly above, placing her face off-center with breathing room along the direction of her gaze. Frame her through softly defocused edges of the playback equipment for a tense, enclosed composition, with directional screen illumination tracing her cheek and tightened brow while the backpack strap remains at the lower edge of frame.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L13B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S20sh2.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L13B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S20sh2.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L13B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L13B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "D", "ranking": [ "D", "A", "B", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "카메라를 모니터 쪽에 위치시켜 인물이 이를 바라보는 공간 논리를 유일하게 지켰으며, 찌푸린 표정과 클로즈업 요구도 훌륭히 충족합니다." }, { "label": "A", "score": 5, "verdict_ko": "표정과 프레이밍은 훌륭하나, 인물이 응시해야 할 메인 모니터 데스크를 등 뒤 배경에 잘못 배치하는 공간 오류가 있습니다." }, { "label": "B", "score": 4, "verdict_ko": "A와 동일하게 모니터 데스크를 배경에 배치한 공간 논리 오류가 있으며, 측면 구도라 정면보다 지시문 구현력이 다소 떨어집 금니다." }, { "label": "C", "score": 3, "verdict_ko": "모니터 화면에 금지된 텍스트가 생성된 하드 위반이 있으며, 메인 데스크를 배경에 배치한 공간 오류도 동일하게 존재합니다." } ] }, "forward_normalized": { "winner": "D", "ranking": [ "D", "A", "B", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "카메라를 모니터 쪽에 위치시켜 인물이 이를 바라보는 공간 논리를 유일하게 지켰으며, 찌푸린 표정과 클로즈업 요구도 훌륭히 충족합니다." }, { "label": "A", "score": 5, "verdict_ko": "표정과 프레이밍은 훌륭하나, 인물이 응시해야 할 메인 모니터 데스크를 등 뒤 배경에 잘못 배치하는 공간 오류가 있습니다." }, { "label": "B", "score": 4, "verdict_ko": "A와 동일하게 모니터 데스크를 배경에 배치한 공간 논리 오류가 있으며, 측면 구도라 정면보다 지시문 구현력이 다소 떨어집 금니다." }, { "label": "C", "score": 3, "verdict_ko": "모니터 화면에 금지된 텍스트가 생성된 하드 위반이 있으며, 메인 데스크를 배경에 배치한 공간 오류도 동일하게 존재합니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'미간을 찌푸린 얼굴 클로즈업' 지시를 충실히 구현했으며 모니터를 향한 시선과 공간 배치가 논리적입니다." }, { "label": "B", "score": 3, "verdict_ko": "우측 모니터 화면에 금지된 텍스트(CCTV)가 유출된 치명적인 오류가 있습니다." }, { "label": "C", "score": 5, "verdict_ko": "얼굴 클로즈업은 좋으나, 상황상 주시해야 할 메인 CCTV 모니터를 등지고 있어 공간 논리가 어색합니다." }, { "label": "D", "score": 4, "verdict_ko": "인물이 메인 CCTV 모니터를 등지고 있어 지시된 상황의 공간적 개연성이 크게 떨어집니다." } ] }, "reverse_normalized": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "'미간을 찌푸린 얼굴 클로즈업' 지시를 충실히 구현했으며 모니터를 향한 시선과 공간 배치가 논리적입니다." }, { "label": "C", "score": 3, "verdict_ko": "우측 모니터 화면에 금지된 텍스트(CCTV)가 유출된 치명적인 오류가 있습니다." }, { "label": "B", "score": 5, "verdict_ko": "얼굴 클로즈업은 좋으나, 상황상 주시해야 할 메인 CCTV 모니터를 등지고 있어 공간 논리가 어색합니다." }, { "label": "A", "score": 4, "verdict_ko": "인물이 메인 CCTV 모니터를 등지고 있어 지시된 상황의 공간적 개연성이 크게 떨어집니다." } ] }, "combined": { "totals": { "A": 9, "B": 9, "C": 6, "D": 14 }, "ranking": [ "D", "A", "B", "C" ], "agreement": true, "policy": 1 } }, "totals": { "A": 9, "B": 9, "C": 6, "D": 14 }, "selected": "D", "ranking": [ "D", "A", "B", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "카메라를 모니터 쪽에 위치시켜 인물이 이를 바라보는 공간 논리를 유일하게 지켰으며, 찌푸린 표정과 클로즈업 요구도 훌륭히 충족합니다." }, { "label": "A", "score": 5, "verdict_ko": "표정과 프레이밍은 훌륭하나, 인물이 응시해야 할 메인 모니터 데스크를 등 뒤 배경에 잘못 배치하는 공간 오류가 있습니다." }, { "label": "B", "score": 4, "verdict_ko": "A와 동일하게 모니터 데스크를 배경에 배치한 공간 논리 오류가 있으며, 측면 구도라 정면보다 지시문 구현력이 다소 떨어집 금니다." }, { "label": "C", "score": 3, "verdict_ko": "모니터 화면에 금지된 텍스트가 생성된 하드 위반이 있으며, 메인 데스크를 배경에 배치한 공간 오류도 동일하게 존재합니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L13B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "카메라가 피사체를 아래에서 올려다보는 구도(로우 앵글)로 되어 있어, 지시된 '약간 위에서(from slightly above)' 촬영해야 한다는 앵글 조건을 위반했습니다.", "fix_en": "Raise the camera to capture the close-up from slightly above her face, rather than looking up from below." }, { "issue_ko": "프레임에 보이는 수리영의 의상이 캐릭터 레퍼런스와 다릅니다(밝은 파란색 재킷과 가슴을 가로지르는 녹색 크로스백 끈이 누락됨).", "fix_en": "Clothe Suri-young in the bright blue zip-up jacket and olive green crossbody strap shown in the character reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Raise the camera to capture the close-up from slightly above her face, rather than looking up from below.\n- Clothe Suri-young in the bright blue zip-up jacket and olive green crossbody strap shown in the character reference.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+엔티티 (4택1: 무콘티 승·D=변형1)" }, "S20sh3::variants": { "author_fp": "8a883201ca2609dd", "author": { "variants": [ { "approach_ko": "CCTV 모니터 화면을 정면 근접 촬영해 비정상적으로 어두운 인물 영상 자체에 시선을 고정한다.", "prompt_en": "Frame the CCTV monitor in a tight, nearly head-on close-up, with its bezel only narrowly visible and the abnormally dark recorded image dominating the still. Keep the man’s complete human figure discernible in posture and clothing while severe underexposure and playback smearing erase identifying detail; use shallow depth so the surrounding office equipment falls softly away. Let the monitor’s subdued glow provide the principal visual emphasis without revealing any interface text." }, { "approach_ko": "보안 사무실 장비 사이에서 모니터를 비스듬히 바라보는 넓은 구도로, 어두운 CCTV 영상의 불길한 이질감을 강조한다.", "prompt_en": "Photograph from a slightly elevated oblique angle in a wider environmental composition, placing the CCTV monitor within layered playback equipment and staff workstations while keeping the screen as the sharp visual anchor. Use deeper focus and restrained early-morning ambient light so the orderly office remains legible, yet the recorded area on the monitor appears conspicuously darker than everything around it. Preserve the man as a complete but unidentifiable human presence within that murky footage, his form degraded by low exposure and image blur rather than reduced to an abstract shape." } ] }, "reused": false }, "S20sh3": { "input_fingerprint": "b95f180e57bcdf75", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 모니터 화면 속, 유독 어둡게 처리되어 형체가 뭉개진 남자의 실루엣이 떠 있는 컷.\n\nLOCATION (lock): A supermarket administrative and security office organized around CCTV monitors, playback equipment, and staff workstations. 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 CCTV recording persistently shows the relevant area as abnormally dark, leaving the man beside Minsuk impossible to identify.\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 모니터 화면 속, 유독 어둡게 처리되어 형체가 뭉개진 남자의 실루엣이 떠 있는 컷.\n\nLOCATION (lock): A supermarket administrative and security office organized around CCTV monitors, playback equipment, and staff workstations. 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 CCTV recording persistently shows the relevant area as abnormally dark, leaving the man beside Minsuk impossible to identify.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame the CCTV monitor in a tight, nearly head-on close-up, with its bezel only narrowly visible and the abnormally dark recorded image dominating the still. Keep the man’s complete human figure discernible in posture and clothing while severe underexposure and playback smearing erase identifying detail; use shallow depth so the surrounding office equipment falls softly away. Let the monitor’s subdued glow provide the principal visual emphasis without revealing any interface text.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 모니터 화면 속, 유독 어둡게 처리되어 형체가 뭉개진 남자의 실루엣이 떠 있는 컷.\n\nLOCATION (lock): A supermarket administrative and security office organized around CCTV monitors, playback equipment, and staff workstations. 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 CCTV recording persistently shows the relevant area as abnormally dark, leaving the man beside Minsuk impossible to identify.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a slightly elevated oblique angle in a wider environmental composition, placing the CCTV monitor within layered playback equipment and staff workstations while keeping the screen as the sharp visual anchor. Use deeper focus and restrained early-morning ambient light so the orderly office remains legible, yet the recorded area on the monitor appears conspicuously darker than everything around it. Preserve the man as a complete but unidentifiable human presence within that murky footage, his form degraded by low exposure and image blur rather than reduced to an abstract shape.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "샷 텍스트가 지시한 대로 모니터 화면에 집중하는 클로즈업 구도를 취했으며, 화면 속 뭉개진 남자의 실루엣을 정확하게 묘사했습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스 이미지의 카메라 구도를 그대로 복사하는 금지 사항을 위반했으며, 모니터 화면을 클로즈업해야 하는 샷의 의도를 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L13B01.png" } ], "critique": { "issues": [ { "issue_ko": "왼쪽 모니터 화면 상단('16:55'), 주 모니터 베젤, 그리고 모니터 아래 노란색 메모지에 텍스트가 포함되어 있어 '텍스트 없음' 및 '인터페이스 텍스트 없음' 지시를 위반했습니다.", "fix_en": "Remove the '16:55' timestamp from the left screen, the text from the main monitor's top-left bezel, and the writing from the yellow sticky note." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the '16:55' timestamp from the left screen, the text from the main monitor's top-left bezel, and the writing from the yellow sticky note.\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.", "ref_mode": "플레이트만 (배경 전용)" }, "S21sh6::variants": { "author_fp": "3f21726158249054", "author": { "variants": [ { "approach_ko": "정면 눈높이의 타이트한 클로즈업과 얕은 심도로 충격에 굳은 미세 표정을 고립시킨다.", "prompt_en": "Photograph her in a tight, eye-level frontal close-up, with the face centered and the crop pressing close around the forehead, cheeks, and chin. Use a gently compressed lens feel and very shallow focus so her widened eyes, parted lips, and arrested facial muscles hold all attention while the police-office depth falls into soft, indistinct texture. Keep the light soft and restrained, preserving natural skin detail and the stunned stillness of the performance." }, { "approach_ko": "비스듬한 근접 앵글과 기울어진 공간선, 더 깊은 배경감으로 갑작스러운 인식의 불안을 강조한다.", "prompt_en": "Take the close-up from a near three-quarter angle at slightly above eye level, placing her face off center as the office geometry recedes diagonally behind her. Use a closer, more immediate lens feel with moderately held depth, allowing recognizable fragments of the surrounding desks, monitors, documents, and enlarged photographs to remain layered behind the sharply focused face without competing with it. Let the existing interior illumination shape one side of her face more firmly than the other, emphasizing the abrupt, frozen realization." } ] }, "reused": false }, "S21sh6": { "input_fingerprint": "6895329bc87a10e9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 크게 뜨고 입을 살짝 벌린 채 굳어버린 얼굴 클로즈업.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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 enlarged photographs of the two mutilated murder victims remain spread across Hye-soo’s desk as Suri-young recognizes their similarity to Minsuk’s apparent wounds.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 크게 뜨고 입을 살짝 벌린 채 굳어버린 얼굴 클로즈업.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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 enlarged photographs of the two mutilated murder victims remain spread across Hye-soo’s desk as Suri-young recognizes their similarity to Minsuk’s apparent wounds.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a tight, eye-level frontal close-up, with the face centered and the crop pressing close around the forehead, cheeks, and chin. Use a gently compressed lens feel and very shallow focus so her widened eyes, parted lips, and arrested facial muscles hold all attention while the police-office depth falls into soft, indistinct texture. Keep the light soft and restrained, preserving natural skin detail and the stunned stillness of the performance.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 크게 뜨고 입을 살짝 벌린 채 굳어버린 얼굴 클로즈업.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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 enlarged photographs of the two mutilated murder victims remain spread across Hye-soo’s desk as Suri-young recognizes their similarity to Minsuk’s apparent wounds.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTake the close-up from a near three-quarter angle at slightly above eye level, placing her face off center as the office geometry recedes diagonally behind her. Use a closer, more immediate lens feel with moderately held depth, allowing recognizable fragments of the surrounding desks, monitors, documents, and enlarged photographs to remain layered behind the sharply focused face without competing with it. Let the existing interior illumination shape one side of her face more firmly than the other, emphasizing the abrupt, frozen realization.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 크게 뜨고 입을 살짝 벌린 채 굳어버린 얼굴 클로즈업.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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 enlarged photographs of the two mutilated murder victims remain spread across Hye-soo’s desk as Suri-young recognizes their similarity to Minsuk’s apparent wounds.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a tight, eye-level frontal close-up, with the face centered and the crop pressing close around the forehead, cheeks, and chin. Use a gently compressed lens feel and very shallow focus so her widened eyes, parted lips, and arrested facial muscles hold all attention while the police-office depth falls into soft, indistinct texture. Keep the light soft and restrained, preserving natural skin detail and the stunned stillness of the performance.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 크게 뜨고 입을 살짝 벌린 채 굳어버린 얼굴 클로즈업.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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 enlarged photographs of the two mutilated murder victims remain spread across Hye-soo’s desk as Suri-young recognizes their similarity to Minsuk’s apparent wounds.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTake the close-up from a near three-quarter angle at slightly above eye level, placing her face off center as the office geometry recedes diagonally behind her. Use a closer, more immediate lens feel with moderately held depth, allowing recognizable fragments of the surrounding desks, monitors, documents, and enlarged photographs to remain layered behind the sharply focused face without competing with it. Let the existing interior illumination shape one side of her face more firmly than the other, emphasizing the abrupt, frozen realization.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S21sh6.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S21sh6.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "C", "ranking": [ "C", "D", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "요구된 얼굴 클로즈업 앵글과 정확한 표정을 구현했으며, 지문에 없는 인물을 철저히 배제하여 가장 우수합니다." }, { "label": "D", "score": 5, "verdict_ko": "인물 제한과 굳어버린 표정은 준수했으나, 지시된 얼굴 클로즈업이 아닌 넓은 미디엄 샷으로 연출되어 순위가 밀렸습니다." }, { "label": "A", "score": 4, "verdict_ko": "클로즈업 프레이밍과 표정 연기는 좋으나, 지문에 언급되지 않은 배경 인물들을 임의로 추가하여 엄격한 인물 제한 규칙을 위반했습니다." }, { "label": "B", "score": 3, "verdict_ko": "프레이밍이 클로즈업을 벗어나 너무 넓으며, 지문에 없는 다수의 경찰 인물들을 배경에 추가하여 중대한 위반을 범했습니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "D", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "요구된 얼굴 클로즈업 앵글과 정확한 표정을 구현했으며, 지문에 없는 인물을 철저히 배제하여 가장 우수합니다." }, { "label": "D", "score": 5, "verdict_ko": "인물 제한과 굳어버린 표정은 준수했으나, 지시된 얼굴 클로즈업이 아닌 넓은 미디엄 샷으로 연출되어 순위가 밀렸습니다." }, { "label": "A", "score": 4, "verdict_ko": "클로즈업 프레이밍과 표정 연기는 좋으나, 지문에 언급되지 않은 배경 인물들을 임의로 추가하여 엄격한 인물 제한 규칙을 위반했습니다." }, { "label": "B", "score": 3, "verdict_ko": "프레이밍이 클로즈업을 벗어나 너무 넓으며, 지문에 없는 다수의 경찰 인물들을 배경에 추가하여 중대한 위반을 범했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A", "D", "C" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지정된 표정 연기와 장소는 잘 표현되었으나, '얼굴 클로즈업' 지시와 달리 상반신과 주변 환경이 많이 포함된 넓은 프레이밍입니다." }, { "label": "B", "score": 7, "verdict_ko": "텍스트가 요구한 '얼굴 클로즈업' 샷 크기와 얼어붙은 표정을 가장 정확히 구현했으며, 지문 외의 인물을 배제하는 규칙을 철저히 준수했습니다." }, { "label": "C", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 경찰관들을 배경에 추가하는 심각한 규정 위반을 범했으며, 프레이밍도 클로즈업 기준보다 넓습니다." }, { "label": "D", "score": 4, "verdict_ko": "프레이밍과 캐릭터의 표정은 매우 적절하나, 지문에 없는 배경 인물들을 임의로 등장시켜 인물 제한 규칙을 크게 위반했습니다." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "D", "A", "B" ], "verdicts": [ { "label": "D", "score": 6, "verdict_ko": "지정된 표정 연기와 장소는 잘 표현되었으나, '얼굴 클로즈업' 지시와 달리 상반신과 주변 환경이 많이 포함된 넓은 프레이밍입니다." }, { "label": "C", "score": 7, "verdict_ko": "텍스트가 요구한 '얼굴 클로즈업' 샷 크기와 얼어붙은 표정을 가장 정확히 구현했으며, 지문 외의 인물을 배제하는 규칙을 철저히 준수했습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 텍스트에 명시되지 않은 경찰관들을 배경에 추가하는 심각한 규정 위반을 범했으며, 프레이밍도 클로즈업 기준보다 넓습니다." }, { "label": "A", "score": 4, "verdict_ko": "프레이밍과 캐릭터의 표정은 매우 적절하나, 지문에 없는 배경 인물들을 임의로 등장시켜 인물 제한 규칙을 크게 위반했습니다." } ] }, "combined": { "totals": { "A": 8, "B": 6, "C": 14, "D": 11 }, "ranking": [ "C", "D", "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 8, "B": 6, "C": 14, "D": 11 }, "selected": "C", "ranking": [ "C", "D", "A", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "요구된 얼굴 클로즈업 앵글과 정확한 표정을 구현했으며, 지문에 없는 인물을 철저히 배제하여 가장 우수합니다." }, { "label": "D", "score": 5, "verdict_ko": "인물 제한과 굳어버린 표정은 준수했으나, 지시된 얼굴 클로즈업이 아닌 넓은 미디엄 샷으로 연출되어 순위가 밀렸습니다." }, { "label": "A", "score": 4, "verdict_ko": "클로즈업 프레이밍과 표정 연기는 좋으나, 지문에 언급되지 않은 배경 인물들을 임의로 추가하여 엄격한 인물 제한 규칙을 위반했습니다." }, { "label": "B", "score": 3, "verdict_ko": "프레이밍이 클로즈업을 벗어나 너무 넓으며, 지문에 없는 다수의 경찰 인물들을 배경에 추가하여 중대한 위반을 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프레임이 지시된 것처럼 이마, 뺨, 턱 주변을 바짝 자르지 않고 머리 전체와 목, 어깨까지 넓게 보여줍니다.", "fix_en": "Zoom in and crop the frame significantly tighter so the edges press closely around her forehead, cheeks, and chin, excluding her neck, shoulders, and the space above her head." }, { "issue_ko": "현재 프레임에 캐릭터가 노출되어 있음에도 불구하고 레퍼런스 이미지의 검은색 헬멧과 파란색 재킷을 착용하고 있지 않습니다.", "fix_en": "Add the black bicycle helmet to her head and the blue jacket over her grey t-shirt to match the character reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Zoom in and crop the frame significantly tighter so the edges press closely around her forehead, cheeks, and chin, excluding her neck, shoulders, and the space above her head.\n- Add the black bicycle helmet to her head and the blue jacket over her grey t-shirt to match the character reference.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+엔티티 (4택1: 무콘티 승·C=변형0)", "plate_select": { "candidates": { "A": "L12B01", "B": "L12B02" }, "assigned": "L12B01", "choice": "A", "confident": true, "reason_ko": "샷 텍스트가 인물의 '얼굴 클로즈업'만을 명시하고 있어 특정 배경이나 서브공간을 구분할 단서가 없으므로 기존 할당을 유지합니다.", "chosen": "L12B01" } }, "S21sh11::variants": { "author_fp": "fbebb0a7dcfaf86b", "author": { "variants": [ { "approach_ko": "눈높이의 중간 와이드 후면 구도로 수리영과 문을 중심축에 두고, 정지한 몸짓과 반쯤 들어간 쪽지를 선명하게 강조한다.", "prompt_en": "Frame a restrained eye-level medium-wide rear view, placing Suri-young’s still figure on a strong axis toward the office doorway. Keep her full posture and the hand paused at her pocket clearly readable, with the half-inserted note visible; use layered desks, monitors, documents, and crime-scene photographs to create controlled depth around her without distracting from the arrested gesture. Hold broad, natural interior clarity with subdued late-morning tonal contrast." }, { "approach_ko": "주머니 높이의 밀착 후면 구도로 반쯤 넣다 멈춘 쪽지와 손을 전경에 크게 두고, 문을 향한 등과 머리는 얕은 심도로 이어 보인다.", "prompt_en": "Photograph from close behind at pocket height, making the stopped hand and half-inserted note the dominant foreground detail while her back and head lead upward toward the office doorway. Use a compressed, shallow-focus composition: the pocket gesture is crisp, her rigid orientation remains unmistakable, and the surrounding police-office clutter falls into soft layered shapes. Let gentle interior light skim the paper edge and hand, concentrating the tension in the unfinished movement." } ] }, "reused": false }, "S21sh11": { "input_fingerprint": "752f2362240f182b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 쪽지를 주머니에 반쯤 밀어넣은 상태로 멈춘 채 사무실 문 밖을 향해 서 있는 수리영의 뒷모습.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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): Suri-young carries the handwritten coordinate note folded inside her pocket as she leaves the police office.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 쪽지를 주머니에 반쯤 밀어넣은 상태로 멈춘 채 사무실 문 밖을 향해 서 있는 수리영의 뒷모습.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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): Suri-young carries the handwritten coordinate note folded inside her pocket as she leaves the police office.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a restrained eye-level medium-wide rear view, placing Suri-young’s still figure on a strong axis toward the office doorway. Keep her full posture and the hand paused at her pocket clearly readable, with the half-inserted note visible; use layered desks, monitors, documents, and crime-scene photographs to create controlled depth around her without distracting from the arrested gesture. Hold broad, natural interior clarity with subdued late-morning tonal contrast.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 쪽지를 주머니에 반쯤 밀어넣은 상태로 멈춘 채 사무실 문 밖을 향해 서 있는 수리영의 뒷모습.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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): Suri-young carries the handwritten coordinate note folded inside her pocket as she leaves the police office.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from close behind at pocket height, making the stopped hand and half-inserted note the dominant foreground detail while her back and head lead upward toward the office doorway. Use a compressed, shallow-focus composition: the pocket gesture is crisp, her rigid orientation remains unmistakable, and the surrounding police-office clutter falls into soft layered shapes. Let gentle interior light skim the paper edge and hand, concentrating the tension in the unfinished movement.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 쪽지를 주머니에 반쯤 밀어넣은 상태로 멈춘 채 사무실 문 밖을 향해 서 있는 수리영의 뒷모습.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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): Suri-young carries the handwritten coordinate note folded inside her pocket as she leaves the police office.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a restrained eye-level medium-wide rear view, placing Suri-young’s still figure on a strong axis toward the office doorway. Keep her full posture and the hand paused at her pocket clearly readable, with the half-inserted note visible; use layered desks, monitors, documents, and crime-scene photographs to create controlled depth around her without distracting from the arrested gesture. Hold broad, natural interior clarity with subdued late-morning tonal contrast.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 쪽지를 주머니에 반쯤 밀어넣은 상태로 멈춘 채 사무실 문 밖을 향해 서 있는 수리영의 뒷모습.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, case documents, enlarged crime-scene photographs, and adjoining access to a meeting area. 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): Suri-young carries the handwritten coordinate note folded inside her pocket as she leaves the police office.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from close behind at pocket height, making the stopped hand and half-inserted note the dominant foreground detail while her back and head lead upward toward the office doorway. Use a compressed, shallow-focus composition: the pocket gesture is crisp, her rigid orientation remains unmistakable, and the surrounding police-office clutter falls into soft layered shapes. Let gentle interior light skim the paper edge and hand, concentrating the tension in the unfinished movement.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S21sh11.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S21sh11.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "오른쪽 허공에 떠 있는 사진 프레임들(콜라주/물리적 오류)이 발생했으며, 금지된 레퍼런스 카메라 구도를 그대로 복사했습니다." }, { "label": "B", "score": 2, "verdict_ko": "왼쪽 팔에 해부학적으로 불가능한 방향의 손(엄지가 바깥쪽을 향함)이 묘사되어 심각한 신체 왜곡 위반입니다." }, { "label": "C", "score": 4, "verdict_ko": "레퍼런스 사진의 카메라 구도와 책상 위 소품 배치를 그대로 복사해 금지 지침을 위반했고 손가락 묘사가 부자연스럽습니다." }, { "label": "D", "score": 7, "verdict_ko": "레퍼런스 구도를 피하는 새로운 카메라 시점을 적용했으며, 인물의 뒷모습과 쪽지를 넣는 동작을 자연스러운 인체로 정확히 구현했습니다." } ] }, "forward_normalized": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "오른쪽 허공에 떠 있는 사진 프레임들(콜라주/물리적 오류)이 발생했으며, 금지된 레퍼런스 카메라 구도를 그대로 복사했습니다." }, { "label": "B", "score": 2, "verdict_ko": "왼쪽 팔에 해부학적으로 불가능한 방향의 손(엄지가 바깥쪽을 향함)이 묘사되어 심각한 신체 왜곡 위반입니다." }, { "label": "C", "score": 4, "verdict_ko": "레퍼런스 사진의 카메라 구도와 책상 위 소품 배치를 그대로 복사해 금지 지침을 위반했고 손가락 묘사가 부자연스럽습니다." }, { "label": "D", "score": 7, "verdict_ko": "레퍼런스 구도를 피하는 새로운 카메라 시점을 적용했으며, 인물의 뒷모습과 쪽지를 넣는 동작을 자연스러운 인체로 정확히 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 샷의 구도, 인물의 의상, 그리고 사무실 배경을 가장 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "배경 묘사는 훌륭하나 지시된 뒷모습 샷에 비해 프레이밍이 불필요하게 넓습니다." }, { "label": "C", "score": 3, "verdict_ko": "필수 의상인 파란색 재킷이 누락되었고, 요구된 뒷모습 전체 구도를 충족하지 못했습니다." }, { "label": "D", "score": 4, "verdict_ko": "구도는 좋으나 우측 책상에 참조 이미지에 없는 세로형 액자들을 임의로 생성한 오류가 있습니다." } ] }, "reverse_normalized": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "지정된 샷의 구도, 인물의 의상, 그리고 사무실 배경을 가장 정확하게 구현했습니다." }, { "label": "C", "score": 6, "verdict_ko": "배경 묘사는 훌륭하나 지시된 뒷모습 샷에 비해 프레이밍이 불필요하게 넓습니다." }, { "label": "B", "score": 3, "verdict_ko": "필수 의상인 파란색 재킷이 누락되었고, 요구된 뒷모습 전체 구도를 충족하지 못했습니다." }, { "label": "A", "score": 4, "verdict_ko": "구도는 좋으나 우측 책상에 참조 이미지에 없는 세로형 액자들을 임의로 생성한 오류가 있습니다." } ] }, "combined": { "totals": { "A": 7, "B": 5, "C": 10, "D": 14 }, "ranking": [ "D", "C", "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 7, "B": 5, "C": 10, "D": 14 }, "selected": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "오른쪽 허공에 떠 있는 사진 프레임들(콜라주/물리적 오류)이 발생했으며, 금지된 레퍼런스 카메라 구도를 그대로 복사했습니다." }, { "label": "B", "score": 2, "verdict_ko": "왼쪽 팔에 해부학적으로 불가능한 방향의 손(엄지가 바깥쪽을 향함)이 묘사되어 심각한 신체 왜곡 위반입니다." }, { "label": "C", "score": 4, "verdict_ko": "레퍼런스 사진의 카메라 구도와 책상 위 소품 배치를 그대로 복사해 금지 지침을 위반했고 손가락 묘사가 부자연스럽습니다." }, { "label": "D", "score": 7, "verdict_ko": "레퍼런스 구도를 피하는 새로운 카메라 시점을 적용했으며, 인물의 뒷모습과 쪽지를 넣는 동작을 자연스러운 인체로 정확히 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L12B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "주머니에 꽂힌 쪽지에 인쇄된 텍스트와 표(그리드)가 뚜렷하게 보여 '텍스트 금지' 규칙을 위반했으며, 접힌 손글씨 쪽지라는 설정과 맞지 않습니다.", "fix_en": "Replace the printed form in the pocket with a plain, slightly folded piece of paper showing no visible text or lines." }, { "issue_ko": "쪽지를 잡고 있는 오른손의 손가락 형태가 종이와 하나로 뭉개져 해부학적으로 비정상적입니다.", "fix_en": "Redraw the right hand holding the note so the fingers are anatomically distinct and naturally positioned." }, { "issue_ko": "캐릭터 레퍼런스에는 없는 두 개의 두꺼운 갈색 스트랩이 바지 뒷주머니 양쪽에 임의로 추가되었습니다.", "fix_en": "Remove the two brown straps hanging from the back pockets of the jeans so the pockets match standard denim." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the printed form in the pocket with a plain, slightly folded piece of paper showing no visible text or lines.\n- Redraw the right hand holding the note so the fingers are anatomically distinct and naturally positioned.\n- Remove the two brown straps hanging from the back pockets of the jeans so the pockets match standard denim.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+엔티티 (4택1: 무콘티 승·D=변형1)", "plate_select": { "candidates": { "A": "L12B01", "B": "L12B02" }, "assigned": "L12B01", "choice": "Candidate A", "confident": true, "reason_ko": "지문에서 요구하는 '사무실 문'이 배경에 명확히 보이는 공간이며, 두 후보가 동일한 공간(서브스페이스)을 보여주므로 규칙에 따라 현재 할당된 후보를 유지합니다.", "kept": "L12B01" } }, "S22sh2::variants": { "author_fp": "90aa839eac0210d2", "author": { "variants": [ { "approach_ko": "수리영의 어깨 너머로 지도 화면과 긴장된 옆얼굴을 함께 잡는 밀착형 관찰 구도.", "prompt_en": "Frame a tight over-the-shoulder medium close shot from slightly above seated eye level, holding Suri-young’s upper body in three-quarter profile while the monitor map occupies the visual foreground. Use restrained depth of field so her fixed gaze and the relevant map area remain legible within the same plane, with the dim café receding softly behind her. Let the monitor provide a cool directional glow across her face, emphasizing concentration and stillness." }, { "approach_ko": "모니터 가장자리를 전경 장벽으로 삼아 정면의 수리영 얼굴과 상체에 집중하는 압박감 있는 구도.", "prompt_en": "Photograph Suri-young from the monitor side in a compressed frontal medium close-up, using the dark back and edge of the monitor as a dominant foreground obstruction while her seated upper body and face remain clearly visible beyond it. Keep the camera near her eye height and isolate her against the subdued rows of workstations, making her unwavering gaze toward the unseen screen the center of the frame. Shape the existing screen light narrowly across her eyes and cheekbones, allowing the surrounding café to fall into deep, natural shadow." } ] }, "reused": false }, "S22sh2": { "input_fingerprint": "e2a5786ee8d5fc31", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두운 PC방 안, 수리영이 모니터 화면의 지도를 응시하며 앉아 있는 상체.\n\nLOCATION (lock): The exterior parking area of a police station leads to a public computer café. Inside, rows of computer desks and a printer provide a workstation for map searching and printing. 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): Suri-young has brought the coordinate note from the police station and uses its coordinates to locate the position on the PC map.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두운 PC방 안, 수리영이 모니터 화면의 지도를 응시하며 앉아 있는 상체.\n\nLOCATION (lock): The exterior parking area of a police station leads to a public computer café. Inside, rows of computer desks and a printer provide a workstation for map searching and printing. 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): Suri-young has brought the coordinate note from the police station and uses its coordinates to locate the position on the PC map.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight over-the-shoulder medium close shot from slightly above seated eye level, holding Suri-young’s upper body in three-quarter profile while the monitor map occupies the visual foreground. Use restrained depth of field so her fixed gaze and the relevant map area remain legible within the same plane, with the dim café receding softly behind her. Let the monitor provide a cool directional glow across her face, emphasizing concentration and stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두운 PC방 안, 수리영이 모니터 화면의 지도를 응시하며 앉아 있는 상체.\n\nLOCATION (lock): The exterior parking area of a police station leads to a public computer café. Inside, rows of computer desks and a printer provide a workstation for map searching and printing. 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): Suri-young has brought the coordinate note from the police station and uses its coordinates to locate the position on the PC map.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph Suri-young from the monitor side in a compressed frontal medium close-up, using the dark back and edge of the monitor as a dominant foreground obstruction while her seated upper body and face remain clearly visible beyond it. Keep the camera near her eye height and isolate her against the subdued rows of workstations, making her unwavering gaze toward the unseen screen the center of the frame. Shape the existing screen light narrowly across her eyes and cheekbones, allowing the surrounding café to fall into deep, natural shadow.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문대로 모니터의 지도와 메모를 잘 구현했으나 지정된 파란색 재킷이 누락되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "화면 양 가장자리에 지문에 없는 불필요한 인물(또는 신체 복제)이 등장하여 치명적인 위반입니다." } ], "refs": [ { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영이 들고 있는 좌표 메모지의 글씨가 적힌 면이 그녀의 눈이 아닌 카메라 쪽을 향하고 있어 소품 방향 규칙을 위반했습니다.", "fix_en": "Flip the note in her hand so its blank back faces the camera and the written side faces Suri-young's eyes." }, { "issue_ko": "모니터 화면의 지도 UI와 메모지에 텍스트가 나타나 있어 '텍스트 없음' 규칙을 위반했습니다.", "fix_en": "Remove all text, letters, and numbers from the computer monitor and the handheld note." }, { "issue_ko": "프레임에 상체가 명확히 보이지만 레퍼런스 이미지에서 착용한 파란색 재킷을 입고 있지 않습니다.", "fix_en": "Add the blue jacket from the reference image over her grey t-shirt." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Flip the note in her hand so its blank back faces the camera and the written side faces Suri-young's eyes.\n- Remove all text, letters, and numbers from the computer monitor and the handheld note.\n- Add the blue jacket from the reference image over her grey t-shirt.\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.", "ref_mode": "플레이트+콘티+엔티티" }, "S22sh3::variants": { "author_fp": "6d4fe2cc3144f761", "author": { "variants": [ { "approach_ko": "책상 바로 위 수직 시점으로 손과 접힌 지도 면을 정돈된 그래픽 구도로 포착한다.", "prompt_en": "Photograph from directly above in a tight overhead close-up, framing only her hands, the half-folded printed maps, a restrained trace of her clothing at the wrists, and the worn desk surface. Use crisp near-field detail and controlled depth so the arrested fold and layered paper edges become the compositional center, with the monitor glow falling softly across the maps against the dark computer-café ambience." }, { "approach_ko": "책상 높이의 비스듬한 측면 클로즈업으로 멈춘 손가락과 겹친 종이의 긴장을 얕은 심도로 강조한다.", "prompt_en": "Shoot from desk height at a close oblique angle, looking across the layered map sheets toward her motionless fingers at the fold. Let the nearest paper edge and fingertips hold sharp focus while the remaining desk and monitor glow dissolve into a dark, shallow background; use low raking light to reveal the paper creases, slight curl, and tactile tension of the interrupted movement." } ] }, "reused": false }, "S22sh3": { "input_fingerprint": "51c08327a8fc5583", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손이 인쇄된 지도 여러 장을 반으로 접은 채 멈춘 클로즈업.\n\nLOCATION (lock): The exterior parking area of a police station leads to a public computer café. Inside, rows of computer desks and a printer provide a workstation for map searching and printing. 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 computer-café lighting, desk surface, monitor glow, printed maps, and the young woman's unchanged clothing. Maintain her seated position at the same workstation. Exclude her face, most of her body, neighboring customers, and unrelated computer equipment from the hand-and-map close-up.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young folds several printed maps for transport; her coordinate note remains with her, and the maps are about to be placed in her doll-adorned 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손이 인쇄된 지도 여러 장을 반으로 접은 채 멈춘 클로즈업.\n\nLOCATION (lock): The exterior parking area of a police station leads to a public computer café. Inside, rows of computer desks and a printer provide a workstation for map searching and printing. 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 computer-café lighting, desk surface, monitor glow, printed maps, and the young woman's unchanged clothing. Maintain her seated position at the same workstation. Exclude her face, most of her body, neighboring customers, and unrelated computer equipment from the hand-and-map close-up.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young folds several printed maps for transport; her coordinate note remains with her, and the maps are about to be placed in her doll-adorned 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from directly above in a tight overhead close-up, framing only her hands, the half-folded printed maps, a restrained trace of her clothing at the wrists, and the worn desk surface. Use crisp near-field detail and controlled depth so the arrested fold and layered paper edges become the compositional center, with the monitor glow falling softly across the maps against the dark computer-café ambience.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손이 인쇄된 지도 여러 장을 반으로 접은 채 멈춘 클로즈업.\n\nLOCATION (lock): The exterior parking area of a police station leads to a public computer café. Inside, rows of computer desks and a printer provide a workstation for map searching and printing. 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 computer-café lighting, desk surface, monitor glow, printed maps, and the young woman's unchanged clothing. Maintain her seated position at the same workstation. Exclude her face, most of her body, neighboring customers, and unrelated computer equipment from the hand-and-map close-up.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young folds several printed maps for transport; her coordinate note remains with her, and the maps are about to be placed in her doll-adorned 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from desk height at a close oblique angle, looking across the layered map sheets toward her motionless fingers at the fold. Let the nearest paper edge and fingertips hold sharp focus while the remaining desk and monitor glow dissolve into a dark, shallow background; use low raking light to reveal the paper creases, slight curl, and tactile tension of the interrupted movement.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 8, "B": 5 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지도를 반으로 접는 동작과 손의 클로즈업 앵글을 정확하게 연출했으며, 의상과 소품의 디테일도 레퍼런스와 잘 일치합니다." }, { "label": "B", "score": 5, "verdict_ko": "지도를 접는 동작 대신 손가락으로 가리키는 포즈를 취하고 있어 지시문과 다릅니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/scene/recipe/S22sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 인쇄된 해상 지도: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "카메라가 수직으로 내려다보는 탑다운 구도가 아니며, 지시된 프레이밍(손과 손목의 옷자락 일부만 포함)을 어기고 상체, 팔 전체, 키보드, 모니터가 프레임에 잡혔습니다.", "fix_en": "Change the camera to a direct top-down overhead angle and tighten the framing to show exclusively the hands, wrists, folded maps, and desk surface, excluding the torso, full arms, monitor, and keyboard." }, { "issue_ko": "텍스트 금지 지시를 위반하여 지도와 아래에 놓인 종이에 글자와 문자들이 선명하게 나타나 있습니다.", "fix_en": "Remove all text, letters, and typography from the printed maps and the piece of paper." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the camera to a direct top-down overhead angle and tighten the framing to show exclusively the hands, wrists, folded maps, and desk surface, excluding the torso, full arms, monitor, and keyboard.\n- Remove all text, letters, and typography from the printed maps and the piece of paper.\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.", "ref_mode": "prev+엔티티" }, "S23sh2::variants": { "author_fp": "dcb78d1b89637b0a", "author": { "variants": [ { "approach_ko": "수리영과 나란히 달리는 밀착 측면 트래킹으로 손의 긴장과 빗속의 속도감을 강조한다.", "prompt_en": "Photograph the moment as a tight side-on tracking shot moving with Suri-young, keeping her profile, forward-pitched torso, and gripping hands sharply dominant in the frame. Use a compressed, intimate lens feel with shallow depth, while the coastal road and sea streak softly behind her; let storm-filtered afternoon light skim across the rain-soaked contours of her face, hair, and clothing." }, { "approach_ko": "낮고 넓은 측면 구도로 폭풍우 치는 해안도로 속 작은 인물의 고투를 환경적으로 보여준다.", "prompt_en": "Photograph the moment from a low, wider side view, allowing the exposed coastal road, sea, and dark storm sky to surround Suri-young and make her forward drive feel hard-won against the weather. Hold a deeper field and strong lateral composition, with the bicycle and rider reading cleanly against the storm-muted background as heavy rain, streaming hair, and trailing clothing carry the motion across the frame." } ] }, "reused": false }, "S23sh2": { "input_fingerprint": "b71c7b905baa44fa", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): afternoon, heavy thunderstorm.\n\nSHOT TEXT (authoritative, Korean): 쏟아지는 빗속에서 자전거 핸들을 꽉 쥐고 상체를 앞으로 숙인 수리영의 측면, 속도에 의해 뒤로 흩날리는 젖은 머리카락과 옷자락.\n\nLOCATION (lock): An exposed coastal road beside the sea, crossed by bicycle through heavy rain beneath dark storm clouds. 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): Suri-young rides her bicycle with the folded printed maps stored in her doll-adorned backpack and the coordinate note still in her possession.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): afternoon, heavy thunderstorm.\n\nSHOT TEXT (authoritative, Korean): 쏟아지는 빗속에서 자전거 핸들을 꽉 쥐고 상체를 앞으로 숙인 수리영의 측면, 속도에 의해 뒤로 흩날리는 젖은 머리카락과 옷자락.\n\nLOCATION (lock): An exposed coastal road beside the sea, crossed by bicycle through heavy rain beneath dark storm clouds. 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): Suri-young rides her bicycle with the folded printed maps stored in her doll-adorned backpack and the coordinate note still in her possession.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a tight side-on tracking shot moving with Suri-young, keeping her profile, forward-pitched torso, and gripping hands sharply dominant in the frame. Use a compressed, intimate lens feel with shallow depth, while the coastal road and sea streak softly behind her; let storm-filtered afternoon light skim across the rain-soaked contours of her face, hair, and clothing.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): afternoon, heavy thunderstorm.\n\nSHOT TEXT (authoritative, Korean): 쏟아지는 빗속에서 자전거 핸들을 꽉 쥐고 상체를 앞으로 숙인 수리영의 측면, 속도에 의해 뒤로 흩날리는 젖은 머리카락과 옷자락.\n\nLOCATION (lock): An exposed coastal road beside the sea, crossed by bicycle through heavy rain beneath dark storm clouds. 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): Suri-young rides her bicycle with the folded printed maps stored in her doll-adorned backpack and the coordinate note still in her possession.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment from a low, wider side view, allowing the exposed coastal road, sea, and dark storm sky to surround Suri-young and make her forward drive feel hard-won against the weather. Hold a deeper field and strong lateral composition, with the bicycle and rider reading cleanly against the storm-muted background as heavy rain, streaming hair, and trailing clothing carry the motion across the frame.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 8, "A": 5 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "스토리보드의 측면 구도와 폭우 내리는 해안가 배경을 잘 구현했으며, 캐릭터 레퍼런스의 지정 복장(파란 바람막이, 헬멧)을 정확히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "구도와 빗속 액션은 훌륭하게 연출되었으나, 캐릭터 레퍼런스의 핵심 복장(파란 바람막이, 헬멧)이 완전히 누락되었습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S23sh2.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "자전거 하단 프레임(다운 튜브)에 텍스트가 포함되어 있어 텍스트 금지 지침을 위반했습니다.", "fix_en": "Remove the white text from the bicycle's down tube." }, { "issue_ko": "왼쪽 팔과 손이 완전히 누락되어 왼쪽 핸들이 빈 상태로 방치되어 있습니다.", "fix_en": "Render the left arm and hand gripping the left side of the handlebar." } ] }, "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 bicycle's down tube.\n- Render the left arm and hand gripping the left side of the handlebar.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S23sh3::variants": { "author_fp": "6a37bbc7fdd42a17", "author": { "variants": [ { "approach_ko": "정면 눈높이의 극도로 밀착된 클로즈업으로 눈빛과 빗물의 정면 돌진감을 강조한다.", "prompt_en": "Photograph her face almost head-on at eye level in an extremely tight close-up, with the eyes commanding the frame beneath the wet fringe and the crop pressing close around forehead, cheeks, and chin. Use a compressed lens feel and very shallow focus so the determined gaze stays crisply legible while rain crossing the focal plane and the storm-dark surroundings dissolve into soft motion and texture. Let the flat, cold overcast light model her soaked face naturally, with tiny wet highlights sharpening the intensity of her forward drive." }, { "approach_ko": "비스듬한 측면의 낮은 밀착 카메라로 젖은 앞머리 너머 한쪽 눈을 포착해 속도와 저항감을 만든다.", "prompt_en": "Take the facial close-up from a close three-quarter angle, slightly below eye level and offset toward the side she is driving into, allowing the wet fringe to layer across the nearer eye without concealing its focus. Favor an intimate, wider lens feel: the near side of her rain-swept face has tactile presence while the far side falls away quickly, creating diagonal tension and a forceful sense of movement through the storm. Keep the background heavily defocused and let diffuse storm light rake gently across the water on her skin, emphasizing resolve through asymmetry rather than frontal confrontation." } ] }, "reused": false }, "S23sh3": { "input_fingerprint": "0c036c97755c279c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): afternoon, heavy thunderstorm.\n\nSHOT TEXT (authoritative, Korean): 수리영의 젖은 앞머리 사이로 결연한 눈빛이 번뜩이는 얼굴 클로즈업.\n\nLOCATION (lock): An exposed coastal road beside the sea, crossed by bicycle through heavy rain beneath dark storm clouds. 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 heavy rain, overcast road lighting, wet hair, wet clothing, and forward-riding posture. Preserve the sense of speed and water streaming across her face. Exclude the bicycle handlebars, most of the road, and her torso from the tight facial close-up.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 doll-adorned backpack containing the printed maps remains with Suri-young as she presses on through the rain.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): afternoon, heavy thunderstorm.\n\nSHOT TEXT (authoritative, Korean): 수리영의 젖은 앞머리 사이로 결연한 눈빛이 번뜩이는 얼굴 클로즈업.\n\nLOCATION (lock): An exposed coastal road beside the sea, crossed by bicycle through heavy rain beneath dark storm clouds. 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 heavy rain, overcast road lighting, wet hair, wet clothing, and forward-riding posture. Preserve the sense of speed and water streaming across her face. Exclude the bicycle handlebars, most of the road, and her torso from the tight facial close-up.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 doll-adorned backpack containing the printed maps remains with Suri-young as she presses on through the rain.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her face almost head-on at eye level in an extremely tight close-up, with the eyes commanding the frame beneath the wet fringe and the crop pressing close around forehead, cheeks, and chin. Use a compressed lens feel and very shallow focus so the determined gaze stays crisply legible while rain crossing the focal plane and the storm-dark surroundings dissolve into soft motion and texture. Let the flat, cold overcast light model her soaked face naturally, with tiny wet highlights sharpening the intensity of her forward drive.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): afternoon, heavy thunderstorm.\n\nSHOT TEXT (authoritative, Korean): 수리영의 젖은 앞머리 사이로 결연한 눈빛이 번뜩이는 얼굴 클로즈업.\n\nLOCATION (lock): An exposed coastal road beside the sea, crossed by bicycle through heavy rain beneath dark storm clouds. 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 heavy rain, overcast road lighting, wet hair, wet clothing, and forward-riding posture. Preserve the sense of speed and water streaming across her face. Exclude the bicycle handlebars, most of the road, and her torso from the tight facial close-up.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 doll-adorned backpack containing the printed maps remains with Suri-young as she presses on through the rain.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTake the facial close-up from a close three-quarter angle, slightly below eye level and offset toward the side she is driving into, allowing the wet fringe to layer across the nearer eye without concealing its focus. Favor an intimate, wider lens feel: the near side of her rain-swept face has tactile presence while the far side falls away quickly, creating diagonal tension and a forceful sense of movement through the storm. Keep the background heavily defocused and let diffuse storm light rake gently across the water on her skin, emphasizing resolve through asymmetry rather than frontal confrontation.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 대로 상반신을 배제한 타이트한 얼굴 클로즈업 프레이밍을 정확히 구현하였으며, 젖은 머리칼과 결연한 눈빛의 묘사가 뛰어납니다." }, { "label": "B", "score": 4, "verdict_ko": "얼굴 클로즈업 지시를 어기고 프레임을 넓혀 상반신과 배낭까지 노출시켰으므로 프레이밍 우선순위에서 크게 감점되었습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S23sh3.png" }, { "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/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/scene/recipe/S23sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "이전 샷에서 확립된 자전거 헬멧과 뺨을 지나는 헬멧 끈이 프레임 내에 보이지 않습니다.", "fix_en": "Add the black bicycle helmet straps running down her cheeks and under her chin, matching her established state in the previous shot." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the black bicycle helmet straps running down her cheeks and under her chin, matching her established state in the previous shot.\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.", "ref_mode": "lane(map_marker): 스케치+prev+엔티티" }, "S24sh4::variants": { "author_fp": "02dd16312cfd7b18", "author": { "variants": [ { "approach_ko": "벽면을 정면으로 압축한 타이트 클로즈업으로 붉은 원과 사진의 불길한 표적성을 강조한다.", "prompt_en": "Photograph the wall almost straight-on in a tight, severe close-up, with the rough red circle dominating the frame and the old photograph held precisely at its visual center. Use a restrained normal-to-short-telephoto feel to flatten the wall into an ominous graphic plane while preserving tactile detail in the paint, bloodstains, paper wear, and wall surface. Keep the dusk-toned illumination dim and directional, with gentle falloff toward the edges and crisp focus across the central evidence." }, { "approach_ko": "방의 어수선한 전경 너머로 벽을 비스듬히 바라보는 넓은 구도로 현장 속 발견의 느낌을 만든다.", "prompt_en": "Frame from an oblique, low-to-eye-level position deeper within the room, using a moderately wide lens feel so the marked bedroom wall sits beyond the disrupted interior in layered depth. Let nearby ransacked elements occupy soft, partial foreground edges while the red circle and centered photograph remain the sharp compositional destination. Use the fading dusk from the open window as subtle lateral ambience, allowing the room to recede into subdued shadow and giving the image the uneasy feeling of evidence discovered within a violated home." } ] }, "reused": false }, "S24sh4": { "input_fingerprint": "801105dea8bbce4c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 방 안쪽 벽면에 거칠게 그려진 붉은색 원과 그 한가운데 꽂힌 낡은 사진 한 장.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with an open living-room window, overturned dining chairs, disturbed household goods, and muddy footprints. A bedroom wall bears a red circle with a bloodstained photograph fixed at its center. 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 apartment remains ransacked, with footprints, an open living-room window, and overturned dining chairs. A red circle is fixed on Suri-young’s bedroom wall with the bloodstained old photograph pinned at its center.\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 방 안쪽 벽면에 거칠게 그려진 붉은색 원과 그 한가운데 꽂힌 낡은 사진 한 장.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with an open living-room window, overturned dining chairs, disturbed household goods, and muddy footprints. A bedroom wall bears a red circle with a bloodstained photograph fixed at its center. 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 apartment remains ransacked, with footprints, an open living-room window, and overturned dining chairs. A red circle is fixed on Suri-young’s bedroom wall with the bloodstained old photograph pinned at its center.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the wall almost straight-on in a tight, severe close-up, with the rough red circle dominating the frame and the old photograph held precisely at its visual center. Use a restrained normal-to-short-telephoto feel to flatten the wall into an ominous graphic plane while preserving tactile detail in the paint, bloodstains, paper wear, and wall surface. Keep the dusk-toned illumination dim and directional, with gentle falloff toward the edges and crisp focus across the central evidence.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 방 안쪽 벽면에 거칠게 그려진 붉은색 원과 그 한가운데 꽂힌 낡은 사진 한 장.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with an open living-room window, overturned dining chairs, disturbed household goods, and muddy footprints. A bedroom wall bears a red circle with a bloodstained photograph fixed at its center. 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 apartment remains ransacked, with footprints, an open living-room window, and overturned dining chairs. A red circle is fixed on Suri-young’s bedroom wall with the bloodstained old photograph pinned at its center.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame from an oblique, low-to-eye-level position deeper within the room, using a moderately wide lens feel so the marked bedroom wall sits beyond the disrupted interior in layered depth. Let nearby ransacked elements occupy soft, partial foreground edges while the red circle and centered photograph remain the sharp compositional destination. Use the fading dusk from the open window as subtle lateral ambience, allowing the room to recede into subdued shadow and giving the image the uneasy feeling of evidence discovered within a violated home.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "샷 텍스트가 강조하는 붉은 원과 피 묻은 낡은 사진에 정확히 집중한 클로즈업 구도를 완벽히 구현하여 가장 높은 우선순위를 충족합니다." }, { "label": "B", "score": 4, "verdict_ko": "지정된 피사체에 집중하지 않고 불필요하게 프레임을 넓혀 거실 전체를 보여주었으며, 아파트의 내부 구조도 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" } ], "critique": { "issues": [ { "issue_ko": "벽 중앙에 꽂힌 물건이 사진이 아니라 아무 내용이 없는 낡은 종이 조각으로 보입니다.", "fix_en": "Replace the blank piece of stained paper in the center with a recognizable old photograph that contains a faded but visible image." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the blank piece of stained paper in the center with a recognizable old photograph that contains a faded but visible image.\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.", "ref_mode": "플레이트만 (배경 전용)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B03", "choice": "Candidate C", "confident": false, "reason_ko": "원문은 '방 안쪽 벽면에 거칠게 그려진 붉은색 원과 그 한가운데 꽂힌 낡은 사진'을 묘사하고 있으나, 제공된 옥탑방 배경 후보(A, B, D) 중 해당 붉은 원이 그려진 침실 벽면을 명확히 보여주는 이미지가 없습니다. 해당 서브공간을 특정할 명확한 시각적 근거가 없으므로 규칙에 따라 현재 할당된 후보를 유지합니다.", "kept": "L04B03" } }, "S24sh6::variants": { "author_fp": "8fd3a17778d1f531", "author": { "variants": [ { "approach_ko": "낡은 사진의 표면과 그 안의 단서를 화면 가득 또렷하게 읽히는 정면 인서트.", "prompt_en": "Photograph this as a tight, nearly frame-filling insert from the inspector’s exact eyeline, with the front of the old photograph held square to the camera while its supporting edges remain outside the crop. Keep the photographic image and its aged, bloodstained surface sharply legible, with only a thin, softly blurred suggestion of the room around it. Use subdued dusk ambience and shallow depth of field to make the photograph feel like the sole piece of evidence under scrutiny." }, { "approach_ko": "사진을 전경의 핵심 단서로 두고 뒤편의 붉은 원과 난장판 공간을 깊이감 있게 연결한 환경적 시점 쇼트.", "prompt_en": "Use a wider subjective view with the old photograph dominating the near foreground, its front angled naturally toward the unseen inspector and camera, while the vacated red circle and the ransacked apartment remain recognizable deeper in the composition. Frame the photograph against the wall so the missing center and the object now under inspection form a strong visual relationship across layered depth. Let cool dusk light from the open window rake across the disorder, keeping the photograph crisp and the background slightly softer but still narratively readable." } ] }, "reused": true }, "S24sh6": { "input_fingerprint": "eff9248ef9535a5c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 7-80년대 식당과 지붕 위 둥근 원, 두 명의 어린 여자아이가 찍힌 낡은 사진의 시점 쇼트.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with an open living-room window, overturned dining chairs, disturbed household goods, and muddy footprints. A bedroom wall bears a red circle with a bloodstained photograph fixed at its center. 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): Suri-young has removed the bloodstained old photograph from the wall circle and holds it for inspection; the ransacked apartment remains unchanged around 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 7-80년대 식당과 지붕 위 둥근 원, 두 명의 어린 여자아이가 찍힌 낡은 사진의 시점 쇼트.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with an open living-room window, overturned dining chairs, disturbed household goods, and muddy footprints. A bedroom wall bears a red circle with a bloodstained photograph fixed at its center. 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): Suri-young has removed the bloodstained old photograph from the wall circle and holds it for inspection; the ransacked apartment remains unchanged around 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph this as a tight, nearly frame-filling insert from the inspector’s exact eyeline, with the front of the old photograph held square to the camera while its supporting edges remain outside the crop. Keep the photographic image and its aged, bloodstained surface sharply legible, with only a thin, softly blurred suggestion of the room around it. Use subdued dusk ambience and shallow depth of field to make the photograph feel like the sole piece of evidence under scrutiny.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 7-80년대 식당과 지붕 위 둥근 원, 두 명의 어린 여자아이가 찍힌 낡은 사진의 시점 쇼트.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with an open living-room window, overturned dining chairs, disturbed household goods, and muddy footprints. A bedroom wall bears a red circle with a bloodstained photograph fixed at its center. 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): Suri-young has removed the bloodstained old photograph from the wall circle and holds it for inspection; the ransacked apartment remains unchanged around 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider subjective view with the old photograph dominating the near foreground, its front angled naturally toward the unseen inspector and camera, while the vacated red circle and the ransacked apartment remain recognizable deeper in the composition. Frame the photograph against the wall so the missing center and the object now under inspection form a strong visual relationship across layered depth. Let cool dusk light from the open window rake across the disorder, keeping the photograph crisp and the background slightly softer but still narratively readable.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 8, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시문이 요구한 사진의 시점 쇼트를 자연스럽게 구현했으며, 사진 속 식당, 지붕 위 원, 두 소녀 및 핏자국 디테일과 난장판이 된 아파트 배경(어스름한 시간대, 진흙 발자국, 벽의 붉은 원)을 모두 정확히 묘사했습니다." }, { "label": "A", "score": 3, "verdict_ko": "사진이 공중에 떠 있는 콜라주 형태로 생성되어 실사 영화 스틸컷 규정(하드 위반)을 어겼으며, 아파트 내부가 충분히 어질러지지 않았고 시간대(해질녘)도 반영되지 않았습니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B07.png" } ], "critique": { "issues": [ { "issue_ko": "인물이나 신체 일부가 보이지 않아야 한다는 지시와 '보이지 않는 조사관'이라는 시각적 접근 지시를 위반하여 프레임 안에 사진을 들고 있는 손이 등장했습니다.", "fix_en": "Remove the hand holding the photograph entirely, leaving the photograph suspended in its exact current position to represent the POV of an unseen inspector." }, { "issue_ko": "어떠한 텍스트도 없어야 한다는 규칙을 위반하여 낡은 사진 속 식당 건물 간판에 한글 텍스트가 있습니다.", "fix_en": "Remove all text and lettering from the restaurant sign inside the old photograph." } ] }, "ref_mode": "플레이트만 (배경 전용)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B03", "choice": "Candidate C", "confident": false, "reason_ko": "원문은 낡은 사진을 보는 시점 쇼트(인서트)를 요구하나, 후보 중 침실 벽의 사진을 클로즈업하거나 해당 시점을 명확히 나타내는 이미지가 없어 기존 할당을 유지함.", "kept": "L04B03" } }, "S25sh1::variants": { "author_fp": "5ddc501ab92e9ae9", "author": { "variants": [ { "approach_ko": "낮고 넓은 원경에서 인물을 작게 두고 안개와 바다의 압도적인 공백을 강조한다.", "prompt_en": "Photograph the moment as a low, distant wide shot, with Suri-young small but clearly human and legible against the expansive harbor and fog-bound sea. Use layered depth from sharply textured wet rocks in the foreground into progressively softer mist, allowing the vast negative space to carry her isolation. Keep the dusk light cool, diffuse, and directionless, with restrained reflections on rain-darkened surfaces and natural atmospheric falloff." }, { "approach_ko": "등 뒤에 밀착한 중근경으로 웅크린 자세와 응시 방향을 강조하고 안개 낀 바다를 압축한다.", "prompt_en": "Photograph the moment from an intimate rear medium distance at roughly her crouched shoulder height, framing her back and compressed posture as the dominant visual anchor while preserving the view she faces. Use a gently compressed lens feel and shallow atmospheric separation, holding tactile detail in her dark hair, clothing, and the nearest wet rock while the harbor dissolves gradually into dense fog. Shape the dim dusk illumination as soft back-skimming ambient light, subtle rather than silhouetting, so her human form remains fully readable." } ] }, "reused": false }, "S25sh1": { "input_fingerprint": "e4f848ae3ee5d1ec", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 자욱한 안개가 깔린 텅 빈 포구, 수리영이 갯바위에 웅크려 앉아 바다를 응시하는 뒷모습.\n\nLOCATION (lock): An empty harbor wrapped in dense fog after a shower, with rocky shoreline edges and no visible boats until a single small vessel emerges through the mist. 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): Suri-young has brought her doll-adorned backpack, the folded maps, and the old bloodstained photograph to the port; the photograph remains in her possession.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 자욱한 안개가 깔린 텅 빈 포구, 수리영이 갯바위에 웅크려 앉아 바다를 응시하는 뒷모습.\n\nLOCATION (lock): An empty harbor wrapped in dense fog after a shower, with rocky shoreline edges and no visible boats until a single small vessel emerges through the mist. 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): Suri-young has brought her doll-adorned backpack, the folded maps, and the old bloodstained photograph to the port; the photograph remains in her possession.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a low, distant wide shot, with Suri-young small but clearly human and legible against the expansive harbor and fog-bound sea. Use layered depth from sharply textured wet rocks in the foreground into progressively softer mist, allowing the vast negative space to carry her isolation. Keep the dusk light cool, diffuse, and directionless, with restrained reflections on rain-darkened surfaces and natural atmospheric falloff.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 자욱한 안개가 깔린 텅 빈 포구, 수리영이 갯바위에 웅크려 앉아 바다를 응시하는 뒷모습.\n\nLOCATION (lock): An empty harbor wrapped in dense fog after a shower, with rocky shoreline edges and no visible boats until a single small vessel emerges through the mist. 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): Suri-young has brought her doll-adorned backpack, the folded maps, and the old bloodstained photograph to the port; the photograph remains in her possession.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment from an intimate rear medium distance at roughly her crouched shoulder height, framing her back and compressed posture as the dominant visual anchor while preserving the view she faces. Use a gently compressed lens feel and shallow atmospheric separation, holding tactile detail in her dark hair, clothing, and the nearest wet rock while the harbor dissolves gradually into dense fog. Shape the dim dusk illumination as soft back-skimming ambient light, subtle rather than silhouetting, so her human form remains fully readable.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 5 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "스토리보드의 구도와 안개 낀 포구의 분위기를 자연스럽게 구현했으나, 인물 레퍼런스의 짧은 머리 대신 긴 머리로 묘사된 점이 아쉽습니다." }, { "label": "B", "score": 5, "verdict_ko": "지도와 피 묻은 사진을 가방 겉면에 물리적으로 불가능한 방식으로 부자연스럽게 덧붙여 현실성이 떨어지며, 헤어스타일 설정도 누락되었습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S25sh1.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물이 캐릭터 레퍼런스의 파란색 재킷과 헬멧을 착용하지 않고, 일치하지 않는 긴 머리와 회색 상의를 입고 있습니다.", "fix_en": "Replace the character's long hair and grey top with the black helmet and blue windbreaker shown in the character reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the character's long hair and grey top with the black helmet and blue windbreaker shown in the character reference.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S25sh2::variants": { "author_fp": "dda6f8d45bd18f7b", "author": { "variants": [ { "approach_ko": "수리영의 무릎과 휴대전화 화면을 수직에 가까운 탑숏으로 밀착해 알림을 즉각적으로 강조한다.", "prompt_en": "Use an intimate, near-overhead close-up aimed squarely at the phone resting over Suri-young’s knees. Keep the screen and missed-call display crisply legible in the compositional center while her dark clothing forms a soft, restrained border around it; let the cool, diffuse dusk light and fog-muted contrast create a hushed, suspended feeling." }, { "approach_ko": "무릎 높이의 비스듬한 측면 클로즈업과 얕은 심도로 휴대전화 알림을 고립시켜 불안한 발견의 순간을 만든다.", "prompt_en": "Photograph from knee height at a tight oblique angle, with the phone screen seen in sharp three-quarter perspective across Suri-young’s knees. Use shallow focus to isolate the missed-call display while the near edge of her clothing falls softly out of focus, preserving only a faint sense of the damp, fogbound harbor atmosphere beyond the close crop; favor gentle side illumination from the muted dusk." } ] }, "reused": false }, "S25sh2": { "input_fingerprint": "363637903c869b6f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 수리영의 무릎 위에서 휴대전화 화면에 혜수의 부재중 전화 알림이 떠 있는 클로즈업.\n\nLOCATION (lock): An empty harbor wrapped in dense fog after a shower, with rocky shoreline edges and no visible boats until a single small vessel emerges through the mist. 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 harbor fog, muted dusk light, seaside moisture, and the young woman's unchanged crouched seated position and clothing. Keep the phone resting over her knees with the missed-call display visible. Exclude her face, the broad harbor view, boats, and most of the rocks from the close-up.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still carries the backpack, maps, and old photograph, while her mobile phone shows several missed calls from Hye-soo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 수리영의 무릎 위에서 휴대전화 화면에 혜수의 부재중 전화 알림이 떠 있는 클로즈업.\n\nLOCATION (lock): An empty harbor wrapped in dense fog after a shower, with rocky shoreline edges and no visible boats until a single small vessel emerges through the mist. 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 harbor fog, muted dusk light, seaside moisture, and the young woman's unchanged crouched seated position and clothing. Keep the phone resting over her knees with the missed-call display visible. Exclude her face, the broad harbor view, boats, and most of the rocks from the close-up.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still carries the backpack, maps, and old photograph, while her mobile phone shows several missed calls from Hye-soo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate, near-overhead close-up aimed squarely at the phone resting over Suri-young’s knees. Keep the screen and missed-call display crisply legible in the compositional center while her dark clothing forms a soft, restrained border around it; let the cool, diffuse dusk light and fog-muted contrast create a hushed, suspended feeling.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 수리영의 무릎 위에서 휴대전화 화면에 혜수의 부재중 전화 알림이 떠 있는 클로즈업.\n\nLOCATION (lock): An empty harbor wrapped in dense fog after a shower, with rocky shoreline edges and no visible boats until a single small vessel emerges through the mist. 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 harbor fog, muted dusk light, seaside moisture, and the young woman's unchanged crouched seated position and clothing. Keep the phone resting over her knees with the missed-call display visible. Exclude her face, the broad harbor view, boats, and most of the rocks from the close-up.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Suri-young still carries the backpack, maps, and old photograph, while her mobile phone shows several missed calls from Hye-soo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from knee height at a tight oblique angle, with the phone screen seen in sharp three-quarter perspective across Suri-young’s knees. Use shallow focus to isolate the missed-call display while the near edge of her clothing falls softly out of focus, preserving only a faint sense of the damp, fogbound harbor atmosphere beyond the close crop; favor gentle side illumination from the muted dusk.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 4, "B": 7 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "이전 샷에서 착용했던 헬멧이 누락되었으며, 프레임에서 제외하도록 지시된 배가 배경에 선명하게 나타나 연속성 및 연출 지시를 위반했습니다." }, { "label": "B", "score": 7, "verdict_ko": "이전 샷의 헬멧과 의상을 정확히 유지했으며, 지시사항대로 넓은 배경과 배를 배제하고 무릎 위 휴대전화에 집중한 훌륭한 클로즈업입니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S25sh2.png" }, { "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/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/scene/recipe/S25sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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": "휴대전화를 양손으로 쥐고 있습니다. 지시사항과 스토리보드에 명시된 대로 휴대전화는 손이 닿지 않은 채 무릎 위에 놓여 있어야 합니다.", "fix_en": "Remove the hands entirely and place the mobile phone resting directly on the knees." }, { "issue_ko": "화면 우측 상단에 인물의 얼굴 하관(턱, 입)이 보입니다. 프롬프트 지시사항에 따라 얼굴은 프레임에서 완전히 제외되어야 합니다.", "fix_en": "Adjust the framing to completely exclude the person's face and helmet from the top right corner." }, { "issue_ko": "휴대전화 화면에 해독할 수 없는 임의의 깨진 문자들이 생성되었습니다.", "fix_en": "Replace the garbled text on the phone screen with the exact Korean missed-call notification UI shown in the prop reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the hands entirely and place the mobile phone resting directly on the knees.\n- Adjust the framing to completely exclude the person's face and helmet from the top right corner.\n- Replace the garbled text on the phone screen with the exact Korean missed-call notification UI shown in the prop reference.\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.", "ref_mode": "lane(map_marker): 스케치+prev+엔티티" }, "S25sh6::variants": { "author_fp": "b9f9709bff9d3d8d", "author": { "variants": [ { "approach_ko": "눈높이의 밀착된 미디엄 클로즈업으로 밧줄을 쥔 손과 무표정한 곁눈질을 한 프레임에 압축한다.", "prompt_en": "Frame an intimate eye-level medium close-up, keeping the rope-gripping hand in the lower foreground while his half-turned face dominates the image. Use a gently compressed lens feel and shallow focus so the restrained sideways glance and near-motionless facial muscles carry the beat, with cool dusk light softened by the dense fog and lingering wetness." }, { "approach_ko": "낮고 넓은 환경적 구도에서 밧줄의 대각선과 안개 속 여백으로 인우의 경계 어린 곁눈질을 강조한다.", "prompt_en": "Photograph the moment as a low, wider environmental medium shot, using the rope as a strong foreground diagonal that leads toward In-woo. Place him off-center with substantial fog-filled negative space in the direction of his glance; retain enough depth to register the wet harbor surroundings while the mist progressively erases the background, making his small head turn feel guarded and deliberate." } ] }, "reused": false }, "S25sh6": { "input_fingerprint": "cbdcebf72ba1b30e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 인우가 밧줄을 쥔 채 고개를 반쯤 돌려 수리영을 힐끗 쳐다보고 있는 무표정한 얼굴.\n\nLOCATION (lock): An empty harbor wrapped in dense fog after a shower, with rocky shoreline edges and no visible boats until a single small vessel emerges through the mist. 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): Suri-young retains the backpack containing her maps and keeps the old photograph with her while asking In-woo for passage.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 인우가 밧줄을 쥔 채 고개를 반쯤 돌려 수리영을 힐끗 쳐다보고 있는 무표정한 얼굴.\n\nLOCATION (lock): An empty harbor wrapped in dense fog after a shower, with rocky shoreline edges and no visible boats until a single small vessel emerges through the mist. 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): Suri-young retains the backpack containing her maps and keeps the old photograph with her while asking In-woo for passage.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame an intimate eye-level medium close-up, keeping the rope-gripping hand in the lower foreground while his half-turned face dominates the image. Use a gently compressed lens feel and shallow focus so the restrained sideways glance and near-motionless facial muscles carry the beat, with cool dusk light softened by the dense fog and lingering wetness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 인우가 밧줄을 쥔 채 고개를 반쯤 돌려 수리영을 힐끗 쳐다보고 있는 무표정한 얼굴.\n\nLOCATION (lock): An empty harbor wrapped in dense fog after a shower, with rocky shoreline edges and no visible boats until a single small vessel emerges through the mist. 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): Suri-young retains the backpack containing her maps and keeps the old photograph with her while asking In-woo for passage.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a low, wider environmental medium shot, using the rope as a strong foreground diagonal that leads toward In-woo. Place him off-center with substantial fog-filled negative space in the direction of his glance; retain enough depth to register the wet harbor surroundings while the mist progressively erases the background, making his small head turn feel guarded and deliberate.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 3, "B": 8 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "스토리보드의 카메라 구도와 선착장 배경을 무시하였으며, 지정된 인물 레퍼런스의 복장(비니, 재킷)을 반영하지 않았습니다." }, { "label": "B", "score": 8, "verdict_ko": "스토리보드의 구도와 배경을 훌륭하게 재현했고, 인물의 복장 레퍼런스는 물론 수리영의 소지품(배낭, 사진)까지 정확하게 묘사했습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S25sh6.png" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영이 들고 있는 사진의 앞면(이미지)이 카메라를 향하고 있습니다. 프롬프트 지시사항에 따라 인물이 보는 방향으로 앞면이 향해야 하므로, 이 구도에서는 사진의 뒷면이 보여야 합니다.", "fix_en": "Turn the photograph in Suri-young's hand around so that the camera sees its blank back instead of the image side." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Turn the photograph in Suri-young's hand around so that the camera sees its blank back instead of the image side.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S26sh1::variants": { "author_fp": "7b29863c7b7062de", "author": { "variants": [ { "approach_ko": "눈높이의 밀도 높은 정면 상반신 숏으로 혜수의 굳은 표정과 통화 중의 정적을 압박감 있게 포착한다.", "prompt_en": "Photograph her in a tight eye-level upper-body frame with a restrained, natural perspective, holding her face and phone in crisp focus. Keep the composition nearly frontal and subtly off-center, letting the dim room recede into soft, layered disorder behind her. Shape the dusk ambience into low-key side light across her mature features, preserving deep, realistic shadow and emphasizing the rigid stillness of her expression." }, { "approach_ko": "비스듬한 측면의 환경적 상반신 숏으로 혜수와 난장판이 된 공간의 깊이를 함께 드러낸다.", "prompt_en": "Use a wider, slightly elevated oblique view while maintaining an upper-body composition, placing her in the near foreground and opening clear depth through the disturbed living space beyond. Let architectural lines and scattered interior elements create an uneasy, unbalanced frame around her rather than isolating her with blur. Use the fading dusk from the open window as a cool directional edge, with the rest of the room falling into textured darkness so her fixed posture feels exposed within the ransacked apartment." } ] }, "reused": false }, "S26sh1": { "input_fingerprint": "7f8dd918fe41e940", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 옥탑방 거실, 혜수가 휴대전화를 귀에 댄 채 굳은 표정으로 서 있는 상체.\n\nLOCATION (lock): A desolate, recently ransacked rooftop apartment with a loose entrance door shifting in the wind and disordered rooms 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 rooftop apartment remains ransacked, with intruder footprints, displaced household items, an open living-room window, and overturned dining chairs; the photograph has been removed from the bedroom wall.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 옥탑방 거실, 혜수가 휴대전화를 귀에 댄 채 굳은 표정으로 서 있는 상체.\n\nLOCATION (lock): A desolate, recently ransacked rooftop apartment with a loose entrance door shifting in the wind and disordered rooms 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 rooftop apartment remains ransacked, with intruder footprints, displaced household items, an open living-room window, and overturned dining chairs; the photograph has been removed from the bedroom wall.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a tight eye-level upper-body frame with a restrained, natural perspective, holding her face and phone in crisp focus. Keep the composition nearly frontal and subtly off-center, letting the dim room recede into soft, layered disorder behind her. Shape the dusk ambience into low-key side light across her mature features, preserving deep, realistic shadow and emphasizing the rigid stillness of her expression.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 옥탑방 거실, 혜수가 휴대전화를 귀에 댄 채 굳은 표정으로 서 있는 상체.\n\nLOCATION (lock): A desolate, recently ransacked rooftop apartment with a loose entrance door shifting in the wind and disordered rooms 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 rooftop apartment remains ransacked, with intruder footprints, displaced household items, an open living-room window, and overturned dining chairs; the photograph has been removed from the bedroom wall.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider, slightly elevated oblique view while maintaining an upper-body composition, placing her in the near foreground and opening clear depth through the disturbed living space beyond. Let architectural lines and scattered interior elements create an uneasy, unbalanced frame around her rather than isolating her with blur. Use the fading dusk from the open window as a cool directional edge, with the rest of the room falling into textured darkness so her fixed posture feels exposed within the ransacked apartment.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 옥탑방 거실, 혜수가 휴대전화를 귀에 댄 채 굳은 표정으로 서 있는 상체.\n\nLOCATION (lock): A desolate, recently ransacked rooftop apartment with a loose entrance door shifting in the wind and disordered rooms 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 rooftop apartment remains ransacked, with intruder footprints, displaced household items, an open living-room window, and overturned dining chairs; the photograph has been removed from the bedroom wall.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a tight eye-level upper-body frame with a restrained, natural perspective, holding her face and phone in crisp focus. Keep the composition nearly frontal and subtly off-center, letting the dim room recede into soft, layered disorder behind her. Shape the dusk ambience into low-key side light across her mature features, preserving deep, realistic shadow and emphasizing the rigid stillness of her expression.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 옥탑방 거실, 혜수가 휴대전화를 귀에 댄 채 굳은 표정으로 서 있는 상체.\n\nLOCATION (lock): A desolate, recently ransacked rooftop apartment with a loose entrance door shifting in the wind and disordered rooms 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 rooftop apartment remains ransacked, with intruder footprints, displaced household items, an open living-room window, and overturned dining chairs; the photograph has been removed from the bedroom wall.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 짙은색 머리, 성숙한 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider, slightly elevated oblique view while maintaining an upper-body composition, placing her in the near foreground and opening clear depth through the disturbed living space beyond. Let architectural lines and scattered interior elements create an uneasy, unbalanced frame around her rather than isolating her with blur. Use the fading dusk from the open window as a cool directional edge, with the rest of the room falling into textured darkness so her fixed posture feels exposed within the ransacked apartment.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S26sh1.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S26sh1.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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", "C", "A", "D" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "상체 프레이밍은 맞으나 현관문 위치에 침실이 배치되어 공간 레퍼런스를 심각하게 위반함." }, { "label": "B", "score": 7, "verdict_ko": "지정된 공간 구조와 상체 프레이밍을 정확히 구현했으며 인물 외형과 의상도 가장 충실히 재현함." }, { "label": "C", "score": 5, "verdict_ko": "프레이밍과 공간 구조는 양호하나 스마트폰 소품(애플 로고)이 레퍼런스와 달라 우선순위에서 밀림." }, { "label": "D", "score": 3, "verdict_ko": "공간 구조가 좌우 반전되었으며 지정된 상체 샷보다 피사체가 넓게 프레이밍되어 감점됨." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "C", "A", "D" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "상체 프레이밍은 맞으나 현관문 위치에 침실이 배치되어 공간 레퍼런스를 심각하게 위반함." }, { "label": "B", "score": 7, "verdict_ko": "지정된 공간 구조와 상체 프레이밍을 정확히 구현했으며 인물 외형과 의상도 가장 충실히 재현함." }, { "label": "C", "score": 5, "verdict_ko": "프레이밍과 공간 구조는 양호하나 스마트폰 소품(애플 로고)이 레퍼런스와 달라 우선순위에서 밀림." }, { "label": "D", "score": 3, "verdict_ko": "공간 구조가 좌우 반전되었으며 지정된 상체 샷보다 피사체가 넓게 프레이밍되어 감점됨." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "D", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "열린 창문과 뒤집힌 식탁 의자 등 이전 샷의 상태를 가장 정확히 유지했으며, 요구된 구도와 인물의 굳은 표정을 훌륭하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "거실 창문이 닫혀 있으며, 프롬프트에 제공된 소품 참조 이미지와 다른 종류의 스마트폰이 묘사되어 감점되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "창문이 닫혀 있으며, 조끼에 지시되지 않은 의미 불명의 텍스트가 생성되어 하드 위반에 해당합니다." }, { "label": "D", "score": 5, "verdict_ko": "인물의 굳은 표정과 상체 구도는 적절하나, 거실 창문이 닫혀 있고 지정된 식탁 의자의 상태가 프레임에 제대로 나타나지 않았습니다." } ] }, "reverse_normalized": { "winner": "D", "ranking": [ "D", "A", "C", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "열린 창문과 뒤집힌 식탁 의자 등 이전 샷의 상태를 가장 정확히 유지했으며, 요구된 구도와 인물의 굳은 표정을 훌륭하게 구현했습니다." }, { "label": "C", "score": 4, "verdict_ko": "거실 창문이 닫혀 있으며, 프롬프트에 제공된 소품 참조 이미지와 다른 종류의 스마트폰이 묘사되어 감점되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "창문이 닫혀 있으며, 조끼에 지시되지 않은 의미 불명의 텍스트가 생성되어 하드 위반에 해당합니다." }, { "label": "A", "score": 5, "verdict_ko": "인물의 굳은 표정과 상체 구도는 적절하나, 거실 창문이 닫혀 있고 지정된 식탁 의자의 상태가 프레임에 제대로 나타나지 않았습니다." } ] }, "combined": { "totals": { "A": 9, "B": 10, "C": 9, "D": 10 }, "ranking": [ "B", "D", "A", "C" ], "agreement": false, "policy": 1 } }, "totals": { "A": 9, "B": 10, "C": 9, "D": 10 }, "selected": "B", "ranking": [ "B", "D", "A", "C" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "상체 프레이밍은 맞으나 현관문 위치에 침실이 배치되어 공간 레퍼런스를 심각하게 위반함." }, { "label": "B", "score": 7, "verdict_ko": "지정된 공간 구조와 상체 프레이밍을 정확히 구현했으며 인물 외형과 의상도 가장 충실히 재현함." }, { "label": "C", "score": 5, "verdict_ko": "프레이밍과 공간 구조는 양호하나 스마트폰 소품(애플 로고)이 레퍼런스와 달라 우선순위에서 밀림." }, { "label": "D", "score": 3, "verdict_ko": "공간 구조가 좌우 반전되었으며 지정된 상체 샷보다 피사체가 넓게 프레이밍되어 감점됨." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" }, { "label": "CHARACTER REFERENCE — 혜수: 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": "거실 창문이 완전히 닫혀 있습니다. 프롬프트는 '열린 거실 창문'을 명시했습니다.", "fix_en": "Open the living room window." }, { "issue_ko": "현관문이 굳게 닫혀 있습니다. 프롬프트는 바람에 흔들리는 느슨하게 열린 문이어야 한다고 지시했습니다.", "fix_en": "Make the entrance door slightly ajar so it appears loose." }, { "issue_ko": "인물의 조끼 가슴 부분에 의미 없는 텍스트가 생성되었습니다. 프롬프트는 모든 형태의 텍스트를 금지합니다.", "fix_en": "Remove the text from the character's vest." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Open the living room window.\n- Make the entrance door slightly ajar so it appears loose.\n- Remove the text from the character's vest.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·B=변형1)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B04", "choice": "Candidate D", "confident": true, "reason_ko": "지문에서 '어두운 옥탑방 거실'을 명시하고 있으며, 현재 할당된 후보가 허름한 옥탑방의 거실 공간을 가장 잘 보여주므로 이를 유지합니다.", "kept": "L04B04" } }, "S26sh4::variants": { "author_fp": "62432dfa036ddf21", "author": { "variants": [ { "approach_ko": "문틈에 바짝 붙은 낮은 시점과 얕은 심도로, 멈춰 선 철문과 그 너머의 어질러진 내부를 긴장감 있게 포착한다.", "prompt_en": "Photograph from very close to the entrance at a low, slightly oblique height, making the narrow opening the dominant vertical line in the frame. Hold sharp focus on the worn metal door edge while the disordered rooms beyond recede into soft, dim dusk depth, creating a tense, restricted glimpse into the deserted interior." }, { "approach_ko": "한발 물러난 정적인 와이드 구도와 깊은 초점으로, 비스듬한 철문과 뒤쪽 방들의 황폐한 공간 관계를 강조한다.", "prompt_en": "Use a restrained wide, static viewpoint from farther back, framing the entrance door diagonally within the surrounding apartment and allowing the rooms beyond to remain clearly legible in deep focus. Let cool dusk ambience fall across the near surfaces while the interior recedes through layered planes, emphasizing the still instant when the wind-shifted door has paused slightly open." } ] }, "reused": false }, "S26sh4": { "input_fingerprint": "c4f022fb6ef0a4ae", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 바람에 흔들려 살짝 열린 채 멈춘 낡은 현관 철문의 틈새 정경.\n\nLOCATION (lock): A desolate, recently ransacked rooftop apartment with a loose entrance door shifting in the wind and disordered rooms 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 apartment remains ransacked and deserted, and its old entrance door is left loose enough to shift in the wind.\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 바람에 흔들려 살짝 열린 채 멈춘 낡은 현관 철문의 틈새 정경.\n\nLOCATION (lock): A desolate, recently ransacked rooftop apartment with a loose entrance door shifting in the wind and disordered rooms 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 apartment remains ransacked and deserted, and its old entrance door is left loose enough to shift in the wind.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from very close to the entrance at a low, slightly oblique height, making the narrow opening the dominant vertical line in the frame. Hold sharp focus on the worn metal door edge while the disordered rooms beyond recede into soft, dim dusk depth, creating a tense, restricted glimpse into the deserted interior.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 바람에 흔들려 살짝 열린 채 멈춘 낡은 현관 철문의 틈새 정경.\n\nLOCATION (lock): A desolate, recently ransacked rooftop apartment with a loose entrance door shifting in the wind and disordered rooms 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 apartment remains ransacked and deserted, and its old entrance door is left loose enough to shift in the wind.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained wide, static viewpoint from farther back, framing the entrance door diagonally within the surrounding apartment and allowing the rooms beyond to remain clearly legible in deep focus. Let cool dusk ambience fall across the near surfaces while the interior recedes through layered planes, emphasizing the still instant when the wind-shifted door has paused slightly open.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "현관문의 틈새로 실내를 들여다보는 정확한 카메라 구도를 구현하여 텍스트의 핵심 연출을 훌륭하게 소화했습니다." }, { "label": "B", "score": 4, "verdict_ko": "문 틈새로 바라보는 시점이라는 명확한 지시를 무시하고 실내 전체를 조망하는 구도를 취해 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L04B02.png" } ], "critique": { "issues": [ { "issue_ko": "철문 가장자리에 초점을 맞추고 배경의 방 내부는 부드럽게 흐려져야(soft depth) 한다는 지침과 달리, 배경의 방 내부가 선명하게 초점이 맞아 있습니다.", "fix_en": "Apply a shallow depth of field to heavily blur the background rooms, keeping only the worn metal door edge in the foreground in sharp focus." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Apply a shallow depth of field to heavily blur the background rooms, keeping only the worn metal door edge in the foreground in sharp focus.\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.", "ref_mode": "플레이트만 (배경 전용)", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B04", "choice": "Candidate D", "confident": true, "reason_ko": "\"낡은 현관 철문\"이라는 텍스트에 부합하는 철제 현관문이 직접적으로 등장하는 서브공간이므로 현재 할당된 이미지를 유지합니다.", "kept": "L04B04" } }, "S27sh4::variants": { "author_fp": "5f323f8f7f6de9a4", "author": { "variants": [ { "approach_ko": "인우의 시점에 가까운 정면 상체 클로즈업으로 수리영의 집요한 눈빛과 사진을 짚은 손을 한 축에 강조한다.", "prompt_en": "Photograph her in an intimate, eye-level upper-body close-up from near In-woo’s eyeline. Keep her sharply focused stare dominant in the upper frame while the old photograph and her pointing finger occupy the lower foreground, creating a direct visual line from the evidence to her face. Use restrained depth of field and soft dusk illumination, with the boat and sea receding unobtrusively behind her." }, { "approach_ko": "비스듬한 측면의 깊이감 있는 상체 구도로 전경의 사진과 손, 중경의 굳은 얼굴, 후경의 바다를 층층이 배치한다.", "prompt_en": "Use a close oblique upper-body composition, placing the photograph and pointing hand nearest the camera while her face sits slightly deeper in the frame, turned into an unwavering off-camera eyeline toward In-woo. Let the wider lens feel preserve spatial tension between hand, face, working deck, wheelhouse, and open sea rather than isolating her. Shape her features with cool lateral dusk light and maintain crisp environmental depth for a taut, confrontational realism." } ] }, "reused": false }, "S27sh4": { "input_fingerprint": "e2ab0e8b17d45100", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영이 낡은 흑백 사진을 손가락으로 짚은 채 인우의 눈을 뚫어지게 응시하는 상체.\n\nLOCATION (lock): The open deck of a small fishing boat far offshore, with a compact wheelhouse and working deck space surrounded by open sea. 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): Suri-young remains aboard with her doll-adorned backpack and printed maps, and she holds out the old photograph while questioning In-woo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영이 낡은 흑백 사진을 손가락으로 짚은 채 인우의 눈을 뚫어지게 응시하는 상체.\n\nLOCATION (lock): The open deck of a small fishing boat far offshore, with a compact wheelhouse and working deck space surrounded by open sea. 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): Suri-young remains aboard with her doll-adorned backpack and printed maps, and she holds out the old photograph while questioning In-woo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in an intimate, eye-level upper-body close-up from near In-woo’s eyeline. Keep her sharply focused stare dominant in the upper frame while the old photograph and her pointing finger occupy the lower foreground, creating a direct visual line from the evidence to her face. Use restrained depth of field and soft dusk illumination, with the boat and sea receding unobtrusively behind her.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영이 낡은 흑백 사진을 손가락으로 짚은 채 인우의 눈을 뚫어지게 응시하는 상체.\n\nLOCATION (lock): The open deck of a small fishing boat far offshore, with a compact wheelhouse and working deck space surrounded by open sea. 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): Suri-young remains aboard with her doll-adorned backpack and printed maps, and she holds out the old photograph while questioning In-woo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close oblique upper-body composition, placing the photograph and pointing hand nearest the camera while her face sits slightly deeper in the frame, turned into an unwavering off-camera eyeline toward In-woo. Let the wider lens feel preserve spatial tension between hand, face, working deck, wheelhouse, and open sea rather than isolating her. Shape her features with cool lateral dusk light and maintain crisp environmental depth for a taut, confrontational realism.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 7, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "수리영이 사진을 짚은 채 상대를 뚫어지게 응시하는 핵심 행동을 훌륭히 구현했으나, 소품 참조 이미지 대신 스토리보드의 스케치를 모방한 점이 아쉽습니다." }, { "label": "A", "score": 3, "verdict_ko": "수리영(우측 인물)이 사진을 가리키며 상대의 눈을 응시해야 하는 가장 중요한 프롬프트 행동 지시를 완전히 위반하고 역할을 반대로 연출했습니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S27sh4.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 1970~1980년대 식당과 두 여자아이가 찍힌 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영이 들고 있는 사진의 내용이 레퍼런스(식당 앞 두 소녀)와 전혀 다른 4명 가족의 그림으로 잘못 묘사되었습니다.", "fix_en": "Change the content of the old photograph to exactly match the reference showing two girls sitting in front of an old building." }, { "issue_ko": "프롬프트의 인물 규정(수리영 외 등장 불가)과 시각적 접근(인우를 향한 카메라 밖 시선)에 위배되게 화면 우측에 남성(인우)이 프레임 안에 등장했습니다.", "fix_en": "Remove the man on the right side of the frame so that Suri-young is looking at an off-camera eyeline." }, { "issue_ko": "수리영의 의상이 캐릭터 레퍼런스의 파란색 재킷과 일치하지 않습니다.", "fix_en": "Change Suri-young's jacket to the bright blue zip-up windbreaker shown in the character reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the content of the old photograph to exactly match the reference showing two girls sitting in front of an old building.\n- Remove the man on the right side of the frame so that Suri-young is looking at an off-camera eyeline.\n- Change Suri-young's jacket to the bright blue zip-up windbreaker shown in the character reference.\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.", "ref_mode": "lane(map_marker): 스케치+엔티티" }, "S27sh8::variants": { "author_fp": "fc201c3088df186e", "author": { "variants": [ { "approach_ko": "조타실 입구를 옆에서 밀착해 팔의 대각선과 두 사람의 표정을 압축하는 긴박한 핸드헬드풍 미디엄 클로즈 투숏.", "prompt_en": "Photograph the confrontation as a tight, shoulder-height side-on medium close two-shot with an immediate handheld feel. Compress both actors against the wheelhouse entrance, using Suri-young’s extended arm as a strong diagonal between their clearly readable faces and opposing body weight. Keep the mid-action crisp, with shallow depth isolating their physical struggle while retaining enough of the doorway to anchor the exact location; shape their faces with directional dusk light from the open deck." }, { "approach_ko": "작업 갑판의 낮은 시점에서 조타실 구조와 바다까지 깊게 담아, 몸싸움을 공간적 대치로 보여주는 절제된 와이드숏.", "prompt_en": "Use a wider, low camera position on the working deck, looking toward the wheelhouse entrance with a restrained, nearly static composition. Let the fixed structure frame the two figures and make their opposing stances and Suri-young’s blocking arm legible in full spatial context, with the open sea held in a separate depth layer beyond the boat. Favor deep focus, firm architectural lines, and cool, even dusk illumination so the tension comes from geometry, distance, and arrested motion rather than camera agitation." } ] }, "reused": false }, "S27sh8": { "input_fingerprint": "932dd6f5730ffeaa", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 입구에서 수리영이 인우의 앞을 가로막으며 뻗은 팔로 밀어내려는 팽팽한 미드액션.\n\nLOCATION (lock): The open deck of a small fishing boat far offshore, with a compact wheelhouse and working deck space surrounded by open sea. 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): Suri-young still has the backpack, maps, and old photograph with her as she blocks In-woo from turning the boat around.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 입구에서 수리영이 인우의 앞을 가로막으며 뻗은 팔로 밀어내려는 팽팽한 미드액션.\n\nLOCATION (lock): The open deck of a small fishing boat far offshore, with a compact wheelhouse and working deck space surrounded by open sea. 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): Suri-young still has the backpack, maps, and old photograph with her as she blocks In-woo from turning the boat around.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the confrontation as a tight, shoulder-height side-on medium close two-shot with an immediate handheld feel. Compress both actors against the wheelhouse entrance, using Suri-young’s extended arm as a strong diagonal between their clearly readable faces and opposing body weight. Keep the mid-action crisp, with shallow depth isolating their physical struggle while retaining enough of the doorway to anchor the exact location; shape their faces with directional dusk light from the open deck.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 입구에서 수리영이 인우의 앞을 가로막으며 뻗은 팔로 밀어내려는 팽팽한 미드액션.\n\nLOCATION (lock): The open deck of a small fishing boat far offshore, with a compact wheelhouse and working deck space surrounded by open sea. 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): Suri-young still has the backpack, maps, and old photograph with her as she blocks In-woo from turning the boat around.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider, low camera position on the working deck, looking toward the wheelhouse entrance with a restrained, nearly static composition. Let the fixed structure frame the two figures and make their opposing stances and Suri-young’s blocking arm legible in full spatial context, with the open sea held in a separate depth layer beyond the boat. Favor deep focus, firm architectural lines, and cool, even dusk illumination so the tension comes from geometry, distance, and arrested motion rather than camera agitation.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 입구에서 수리영이 인우의 앞을 가로막으며 뻗은 팔로 밀어내려는 팽팽한 미드액션.\n\nLOCATION (lock): The open deck of a small fishing boat far offshore, with a compact wheelhouse and working deck space surrounded by open sea. 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): Suri-young still has the backpack, maps, and old photograph with her as she blocks In-woo from turning the boat around.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the confrontation as a tight, shoulder-height side-on medium close two-shot with an immediate handheld feel. Compress both actors against the wheelhouse entrance, using Suri-young’s extended arm as a strong diagonal between their clearly readable faces and opposing body weight. Keep the mid-action crisp, with shallow depth isolating their physical struggle while retaining enough of the doorway to anchor the exact location; shape their faces with directional dusk light from the open deck.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 입구에서 수리영이 인우의 앞을 가로막으며 뻗은 팔로 밀어내려는 팽팽한 미드액션.\n\nLOCATION (lock): The open deck of a small fishing boat far offshore, with a compact wheelhouse and working deck space surrounded by open sea. 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): Suri-young still has the backpack, maps, and old photograph with her as she blocks In-woo from turning the boat around.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider, low camera position on the working deck, looking toward the wheelhouse entrance with a restrained, nearly static composition. Let the fixed structure frame the two figures and make their opposing stances and Suri-young’s blocking arm legible in full spatial context, with the open sea held in a separate depth layer beyond the boat. Favor deep focus, firm architectural lines, and cool, even dusk illumination so the tension comes from geometry, distance, and arrested motion rather than camera agitation.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L18B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_fishing_boat_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S27sh8.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 인우: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L18B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_fishing_boat_sel.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S27sh8.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L18B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_fishing_boat_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L18B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_fishing_boat_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "C", "ranking": [ "C", "B", "D", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 미드샷 구도와 상대를 밀어내는 팽팽한 액션을 가장 정확히 구현했으며, 복장과 소지품 조건도 잘 충족했습니다." }, { "label": "B", "score": 5, "verdict_ko": "밀어내는 액션과 소지품은 명확하지만, 구도가 지시된 미드샷보다 넓으며 인우의 비니가 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "수리영이 인우가 아닌 조타실 기기를 밀고 있어 상대를 가로막는다는 핵심 액션 지시를 벗어났습니다." }, { "label": "A", "score": 3, "verdict_ko": "두 인물이 같은 방향을 보며 서 있어, 인우의 앞을 가로막고 밀어낸다는 지시를 완전히 위반했습니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "B", "D", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 미드샷 구도와 상대를 밀어내는 팽팽한 액션을 가장 정확히 구현했으며, 복장과 소지품 조건도 잘 충족했습니다." }, { "label": "B", "score": 5, "verdict_ko": "밀어내는 액션과 소지품은 명확하지만, 구도가 지시된 미드샷보다 넓으며 인우의 비니가 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "수리영이 인우가 아닌 조타실 기기를 밀고 있어 상대를 가로막는다는 핵심 액션 지시를 벗어났습니다." }, { "label": "A", "score": 3, "verdict_ko": "두 인물이 같은 방향을 보며 서 있어, 인우의 앞을 가로막고 밀어낸다는 지시를 완전히 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "C", "A", "D" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "소품과 수리영의 단발은 정확하나, 거리가 멀어 가로막고 밀어내는 팽팽한 액션의 긴장감이 부족합니다." }, { "label": "B", "score": 7, "verdict_ko": "가로막고 밀어내는 팽팽한 액션과 인우의 비니 착용을 잘 구현했으나, 수리영의 단발머리가 묶여 있는 점이 아쉽습니다." }, { "label": "C", "score": 6, "verdict_ko": "액션과 소품 구성은 적절하나, 인우의 비니가 누락되었고 레퍼런스 대비 얼굴 일치도가 떨어집니다." }, { "label": "D", "score": 3, "verdict_ko": "앞을 가로막고 밀어내는 지문을 완전히 위반하고, 두 사람이 나란히 서서 같은 방향을 응시하여 오답입니다." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "B", "D", "A" ], "verdicts": [ { "label": "D", "score": 5, "verdict_ko": "소품과 수리영의 단발은 정확하나, 거리가 멀어 가로막고 밀어내는 팽팽한 액션의 긴장감이 부족합니다." }, { "label": "C", "score": 7, "verdict_ko": "가로막고 밀어내는 팽팽한 액션과 인우의 비니 착용을 잘 구현했으나, 수리영의 단발머리가 묶여 있는 점이 아쉽습니다." }, { "label": "B", "score": 6, "verdict_ko": "액션과 소품 구성은 적절하나, 인우의 비니가 누락되었고 레퍼런스 대비 얼굴 일치도가 떨어집니다." }, { "label": "A", "score": 3, "verdict_ko": "앞을 가로막고 밀어내는 지문을 완전히 위반하고, 두 사람이 나란히 서서 같은 방향을 응시하여 오답입니다." } ] }, "combined": { "totals": { "A": 6, "B": 11, "C": 14, "D": 9 }, "ranking": [ "C", "B", "D", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 11, "C": 14, "D": 9 }, "selected": "C", "ranking": [ "C", "B", "D", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 미드샷 구도와 상대를 밀어내는 팽팽한 액션을 가장 정확히 구현했으며, 복장과 소지품 조건도 잘 충족했습니다." }, { "label": "B", "score": 5, "verdict_ko": "밀어내는 액션과 소지품은 명확하지만, 구도가 지시된 미드샷보다 넓으며 인우의 비니가 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "수리영이 인우가 아닌 조타실 기기를 밀고 있어 상대를 가로막는다는 핵심 액션 지시를 벗어났습니다." }, { "label": "A", "score": 3, "verdict_ko": "두 인물이 같은 방향을 보며 서 있어, 인우의 앞을 가로막고 밀어낸다는 지시를 완전히 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L18B01.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/61621f50-9d47-4fdc-9076-02a839c1733c/images/background_chain/seed_bg_fishing_boat_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "조타실 구조물과 문이 'Structure Look' 사진(흰색 벽, 파란색 문)이 아닌 'Location' 사진의 회색 금속 디자인으로 렌더링되었습니다.", "fix_en": "Change the wheelhouse structure and door to match the white paint and blue panels shown in the Structure Look photograph." }, { "issue_ko": "수리영의 복장이 캐릭터 레퍼런스(파란색 재킷, 검은색 헬멧)와 일치하지 않습니다.", "fix_en": "Change Suri-young's clothing to the blue jacket and black helmet from her character reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the wheelhouse structure and door to match the white paint and blue panels shown in the Structure Look photograph.\n- Change Suri-young's clothing to the blue jacket and black helmet from her character reference.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+seed+엔티티 (복잡구조물 4택1: 무콘티 승·C=변형0)", "lane_policy": "ab_select_ready", "plate_select": { "candidates": { "A": "L18B01", "B": "L18B02" }, "assigned": "L18B01", "choice": "Candidate A", "confident": true, "reason_ko": "\"조타실 입구에서\"라는 지문에 맞춰 조타실 문이 열려 있는 갑판 입구를 정확히 보여주는 이미지입니다. 두 후보가 동일하므로 현재 배정된 A를 유지합니다.", "kept": "L18B01" } }, "S28sh3::variants": { "author_fp": "8d82a4e0c2018f10", "author": { "variants": [ { "approach_ko": "사진을 거의 수직으로 내려다보며 손가락과 눌린 얼굴을 평면적이고 도상적으로 강조한 극근접 구도.", "prompt_en": "Photograph the moment in an almost perpendicular overhead close-up, with the old photograph filling nearly the entire frame. Use a restrained, natural perspective and precise focus on the fingertip and the girl's partially obscured face, allowing the photograph's edges and surrounding wheelhouse surfaces to fall softly out of focus. Cool dusk light from the windows gives the image a quiet, forensic stillness." }, { "approach_ko": "사진 표면과 나란한 낮은 사선에서 압력과 종이의 질감을 촉각적으로 드러내는 밀착 구도.", "prompt_en": "Shoot from a low, tight oblique angle nearly level with the photograph, making the downward pressure of the finger the dominant physical gesture. Let shallow focus travel across the fingertip into the girl's face while raking dusk light reveals the worn paper texture; dissolve the steering controls and circular interior markings into a compressed, dim background. The framing feels intimate and emotionally charged rather than evidentiary." } ] }, "reused": false }, "S28sh3": { "input_fingerprint": "f1ba550516c53dd5", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손가락이 낡은 사진 속 한 여자아이의 얼굴을 꾹 누르고 있는 클로즈업.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Suri-young continues holding the old photograph and presses a finger to the girl she identifies as her mother; her backpack and maps remain aboard with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손가락이 낡은 사진 속 한 여자아이의 얼굴을 꾹 누르고 있는 클로즈업.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Suri-young continues holding the old photograph and presses a finger to the girl she identifies as her mother; her backpack and maps remain aboard with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in an almost perpendicular overhead close-up, with the old photograph filling nearly the entire frame. Use a restrained, natural perspective and precise focus on the fingertip and the girl's partially obscured face, allowing the photograph's edges and surrounding wheelhouse surfaces to fall softly out of focus. Cool dusk light from the windows gives the image a quiet, forensic stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손가락이 낡은 사진 속 한 여자아이의 얼굴을 꾹 누르고 있는 클로즈업.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Suri-young continues holding the old photograph and presses a finger to the girl she identifies as her mother; her backpack and maps remain aboard with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from a low, tight oblique angle nearly level with the photograph, making the downward pressure of the finger the dominant physical gesture. Let shallow focus travel across the fingertip into the girl's face while raking dusk light reveals the worn paper texture; dissolve the steering controls and circular interior markings into a compressed, dim background. The framing feels intimate and emotionally charged rather than evidentiary.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손가락이 낡은 사진 속 한 여자아이의 얼굴을 꾹 누르고 있는 클로즈업.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Suri-young continues holding the old photograph and presses a finger to the girl she identifies as her mother; her backpack and maps remain aboard with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in an almost perpendicular overhead close-up, with the old photograph filling nearly the entire frame. Use a restrained, natural perspective and precise focus on the fingertip and the girl's partially obscured face, allowing the photograph's edges and surrounding wheelhouse surfaces to fall softly out of focus. Cool dusk light from the windows gives the image a quiet, forensic stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손가락이 낡은 사진 속 한 여자아이의 얼굴을 꾹 누르고 있는 클로즈업.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Suri-young continues holding the old photograph and presses a finger to the girl she identifies as her mother; her backpack and maps remain aboard with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from a low, tight oblique angle nearly level with the photograph, making the downward pressure of the finger the dominant physical gesture. Let shallow focus travel across the fingertip into the girl's face while raking dusk light reveals the worn paper texture; dissolve the steering controls and circular interior markings into a compressed, dim background. The framing feels intimate and emotionally charged rather than evidentiary.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S28sh3.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 1970~1980년대 식당과 두 여자아이가 찍힌 낡은 사진: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S28sh3.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 1970~1980년대 식당과 두 여자아이가 찍힌 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 1970~1980년대 식당과 두 여자아이가 찍힌 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 1970~1980년대 식당과 두 여자아이가 찍힌 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "C", "ranking": [ "C", "A", "D", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 클로즈업 샷과 손가락으로 사진 속 얼굴을 누르는 동작, 정확한 사진 소품과 배경 및 의상(파란 재킷)을 가장 완벽하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "클로즈업의 의도와 누르는 동작은 표현되었으나, 손이 위에서 내려오는 구도가 어색하며 사진을 들고 있는 상태가 명확하지 않습니다." }, { "label": "D", "score": 4, "verdict_ko": "기준 사진 소품의 디테일이 심하게 왜곡되어 뭉개졌으며, 손에 들고 누르기보다 바닥에 두고 가리키는 구도에 가깝습니다." }, { "label": "B", "score": 3, "verdict_ko": "클로즈업 지시를 어기고 프레임을 넓혀 인물 얼굴을 포함시켰으며, 기준과 전혀 다른 잘못된 내용의 사진 소품을 생성했습니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "A", "D", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 클로즈업 샷과 손가락으로 사진 속 얼굴을 누르는 동작, 정확한 사진 소품과 배경 및 의상(파란 재킷)을 가장 완벽하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "클로즈업의 의도와 누르는 동작은 표현되었으나, 손이 위에서 내려오는 구도가 어색하며 사진을 들고 있는 상태가 명확하지 않습니다." }, { "label": "D", "score": 4, "verdict_ko": "기준 사진 소품의 디테일이 심하게 왜곡되어 뭉개졌으며, 손에 들고 누르기보다 바닥에 두고 가리키는 구도에 가깝습니다." }, { "label": "B", "score": 3, "verdict_ko": "클로즈업 지시를 어기고 프레임을 넓혀 인물 얼굴을 포함시켰으며, 기준과 전혀 다른 잘못된 내용의 사진 소품을 생성했습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "지시된 정확한 클로즈업 프레이밍을 사용했으며, 손가락이 사진 속 소녀의 얼굴을 꾹 누르는 액션과 소품을 완벽히 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "손가락이 얼굴을 향해 있으나 꾹 누르는 대신 단순히 가리키고 있으며, 요구된 클로즈업보다 샷이 너무 넓습니다." }, { "label": "A", "score": 4, "verdict_ko": "클로즈업 샷이 아니며, 손가락이 소녀의 얼굴이 아닌 사진 속 건물의 지붕을 가리켜 핵심 액션을 실패했습니다." }, { "label": "C", "score": 3, "verdict_ko": "제공된 레퍼런스와 완전히 다른 인물 중심의 사진을 소품으로 사용하여 규정을 심각하게 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 정확한 클로즈업 프레이밍을 사용했으며, 손가락이 사진 속 소녀의 얼굴을 꾹 누르는 액션과 소품을 완벽히 구현했습니다." }, { "label": "C", "score": 6, "verdict_ko": "손가락이 얼굴을 향해 있으나 꾹 누르는 대신 단순히 가리키고 있으며, 요구된 클로즈업보다 샷이 너무 넓습니다." }, { "label": "D", "score": 4, "verdict_ko": "클로즈업 샷이 아니며, 손가락이 소녀의 얼굴이 아닌 사진 속 건물의 지붕을 가리켜 핵심 액션을 실패했습니다." }, { "label": "B", "score": 3, "verdict_ko": "제공된 레퍼런스와 완전히 다른 인물 중심의 사진을 소품으로 사용하여 규정을 심각하게 위반했습니다." } ] }, "combined": { "totals": { "A": 12, "B": 6, "C": 13, "D": 8 }, "ranking": [ "C", "A", "D", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 12, "B": 6, "C": 13, "D": 8 }, "selected": "C", "ranking": [ "C", "A", "D", "B" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 클로즈업 샷과 손가락으로 사진 속 얼굴을 누르는 동작, 정확한 사진 소품과 배경 및 의상(파란 재킷)을 가장 완벽하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "클로즈업의 의도와 누르는 동작은 표현되었으나, 손이 위에서 내려오는 구도가 어색하며 사진을 들고 있는 상태가 명확하지 않습니다." }, { "label": "D", "score": 4, "verdict_ko": "기준 사진 소품의 디테일이 심하게 왜곡되어 뭉개졌으며, 손에 들고 누르기보다 바닥에 두고 가리키는 구도에 가깝습니다." }, { "label": "B", "score": 3, "verdict_ko": "클로즈업 지시를 어기고 프레임을 넓혀 인물 얼굴을 포함시켰으며, 기준과 전혀 다른 잘못된 내용의 사진 소품을 생성했습니다." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 1970~1980년대 식당과 두 여자아이가 찍힌 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "샷 텍스트에는 손가락이 여자아이의 얼굴을 꾹 누르고 있어야 하지만, 생성된 이미지에서는 손가락이 건물 벽을 가리키고 있습니다.", "fix_en": "Reposition the right index finger so its tip firmly presses down directly onto the face of the girl sitting on the left in the photograph." }, { "issue_ko": "'텍스트 없음' 규칙을 위반하여 낡은 사진 속 건물 벽(여자아이들 옆)에 의미를 알 수 없는 문자('hm... 920%')가 생성되었습니다.", "fix_en": "Erase the gibberish text from the wall of the building in the photograph, replacing it with plain wall texture." }, { "issue_ko": "사진을 가리키는 오른손의 엄지손가락이 손목 아래에 분리된 형태처럼 비정상적으로 붙어 있어 해부학적으로 틀렸습니다.", "fix_en": "Redraw the right hand to correct its anatomy, ensuring the thumb connects naturally to the side of the palm rather than protruding from under the wrist." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Reposition the right index finger so its tip firmly presses down directly onto the face of the girl sitting on the left in the photograph.\n- Erase the gibberish text from the wall of the building in the photograph, replacing it with plain wall texture.\n- Redraw the right hand to correct its anatomy, ensuring the thumb connects naturally to the side of the palm rather than protruding from under the wrist.\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.", "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": false, "ref_mode": "플레이트+엔티티 (4택1: 무콘티 승·C=변형0)" }, "S28sh5::variants": { "author_fp": "34836c28b660d5e4", "author": { "variants": [ { "approach_ko": "인우의 어깨를 크게 전경에 둔 밀착형 오버숄더로, 얕은 심도 속 표식 벽면에 시선을 집중한다.", "prompt_en": "Frame a tight, intimate over-the-shoulder view from just behind Inwoo, with his neatly groomed dark hair and shoulder occupying a substantial soft foreground edge. Let the densely marked wall dominate the plane of focus, using restrained shallow depth and a slightly compressed lens feel to make the strange circles seem visually crowded. Cool dusk ambience falls gently across the surfaces, with subdued contrast and tactile live-action texture." }, { "approach_ko": "조금 높고 넓은 오버숄더 구도로 조타실의 깊이와 벽면 전체의 표식 밀도를 함께 드러낸다.", "prompt_en": "Photograph the moment as a wider over-the-shoulder composition from slightly above Inwoo’s shoulder line, keeping him clearly readable from behind while revealing the cramped spatial depth of the wheelhouse. Use a natural wide-angle feel and deeper focus so the steering area, windows, sea beyond, and the circular markings across the interior surfaces remain spatially coherent, with the marked wall carrying the strongest compositional weight. Shape the dusk illumination as soft directional falloff from the windows, allowing the enclosed interior to feel tense without obscuring physical detail." } ] }, "reused": false }, "S28sh5": { "input_fingerprint": "096e4e8a17cedf9b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 인우의 어깨 너머로 조타실 벽면에 빼곡하게 그려진 기괴한 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Numerous circular symbols remain drawn throughout the wheelhouse, including the newly noticed red circle. Suri-young still has the old photograph and her backpack of maps with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 인우의 어깨 너머로 조타실 벽면에 빼곡하게 그려진 기괴한 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Numerous circular symbols remain drawn throughout the wheelhouse, including the newly noticed red circle. Suri-young still has the old photograph and her backpack of maps with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight, intimate over-the-shoulder view from just behind Inwoo, with his neatly groomed dark hair and shoulder occupying a substantial soft foreground edge. Let the densely marked wall dominate the plane of focus, using restrained shallow depth and a slightly compressed lens feel to make the strange circles seem visually crowded. Cool dusk ambience falls gently across the surfaces, with subdued contrast and tactile live-action texture.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 인우의 어깨 너머로 조타실 벽면에 빼곡하게 그려진 기괴한 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Numerous circular symbols remain drawn throughout the wheelhouse, including the newly noticed red circle. Suri-young still has the old photograph and her backpack of maps with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a wider over-the-shoulder composition from slightly above Inwoo’s shoulder line, keeping him clearly readable from behind while revealing the cramped spatial depth of the wheelhouse. Use a natural wide-angle feel and deeper focus so the steering area, windows, sea beyond, and the circular markings across the interior surfaces remain spatially coherent, with the marked wall carrying the strongest compositional weight. Shape the dusk illumination as soft directional falloff from the windows, allowing the enclosed interior to feel tense without obscuring physical detail.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 인우의 어깨 너머로 조타실 벽면에 빼곡하게 그려진 기괴한 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Numerous circular symbols remain drawn throughout the wheelhouse, including the newly noticed red circle. Suri-young still has the old photograph and her backpack of maps with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight, intimate over-the-shoulder view from just behind Inwoo, with his neatly groomed dark hair and shoulder occupying a substantial soft foreground edge. Let the densely marked wall dominate the plane of focus, using restrained shallow depth and a slightly compressed lens feel to make the strange circles seem visually crowded. Cool dusk ambience falls gently across the surfaces, with subdued contrast and tactile live-action texture.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 인우의 어깨 너머로 조타실 벽면에 빼곡하게 그려진 기괴한 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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): Numerous circular symbols remain drawn throughout the wheelhouse, including the newly noticed red circle. Suri-young still has the old photograph and her backpack of maps with her.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a wider over-the-shoulder composition from slightly above Inwoo’s shoulder line, keeping him clearly readable from behind while revealing the cramped spatial depth of the wheelhouse. Use a natural wide-angle feel and deeper focus so the steering area, windows, sea beyond, and the circular markings across the interior surfaces remain spatially coherent, with the marked wall carrying the strongest compositional weight. Shape the dusk illumination as soft directional falloff from the windows, allowing the enclosed interior to feel tense without obscuring physical detail.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "roll_refs": { "A": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S28sh5.png" }, { "label": "CHARACTER REFERENCE — 인우: 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "LAYOUT SKETCH — a bare thin-line layout guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, pose and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic reference. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/conti_S28sh5.png" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "C": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "D": [ { "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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "어깨를 전경에 크게 걸쳐 우측 벽면의 원형 표식들을 강조하는 '어깨 너머 시점 쇼트'의 카메라 구도를 가장 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "의상과 로케이션 일치도는 매우 높으나, 인물이 프레임 안쪽으로 들어가 있어 지정된 오버더숄더 구도보다는 일반적인 후면 쇼트에 가깝습니다." }, { "label": "C", "score": 3, "verdict_ko": "레퍼런스의 조타실 콘솔 디자인과 우측 벽면/출입구 구조를 임의로 왜곡하여 로케이션 고정 지시를 위반했습니다." }, { "label": "D", "score": 4, "verdict_ko": "인물이 조타석에 앉아 정면을 향하는 넓은 후면 쇼트로 연출되어, 벽면 표식을 향한 어깨 너머 시점 쇼트라는 핵심 구도를 벗어났습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "어깨를 전경에 크게 걸쳐 우측 벽면의 원형 표식들을 강조하는 '어깨 너머 시점 쇼트'의 카메라 구도를 가장 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "의상과 로케이션 일치도는 매우 높으나, 인물이 프레임 안쪽으로 들어가 있어 지정된 오버더숄더 구도보다는 일반적인 후면 쇼트에 가깝습니다." }, { "label": "C", "score": 3, "verdict_ko": "레퍼런스의 조타실 콘솔 디자인과 우측 벽면/출입구 구조를 임의로 왜곡하여 로케이션 고정 지시를 위반했습니다." }, { "label": "D", "score": 4, "verdict_ko": "인물이 조타석에 앉아 정면을 향하는 넓은 후면 쇼트로 연출되어, 벽면 표식을 향한 어깨 너머 시점 쇼트라는 핵심 구도를 벗어났습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "정확한 오버더숄더 시점 쇼트 구도로 벽면의 표식에 완벽히 초점을 맞추었으나, 레퍼런스의 비니가 누락되었습니다." }, { "label": "B", "score": 6, "verdict_ko": "오버더숄더 구도는 적절하나 전경 좌측에 식별하기 어려운 회색 물체가 시야를 일부 가립니다." }, { "label": "A", "score": 5, "verdict_ko": "레퍼런스의 의상(비니)을 잘 재현했으나, 앵글이 다소 넓어 몰입감 있는 시점 쇼트로 보기 어렵습니다." }, { "label": "C", "score": 4, "verdict_ko": "카메라 구도가 미디엄 쇼트에 가까워 요구된 오버더숄더 시점 쇼트의 조건을 충족하지 못합니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "정확한 오버더숄더 시점 쇼트 구도로 벽면의 표식에 완벽히 초점을 맞추었으나, 레퍼런스의 비니가 누락되었습니다." }, { "label": "C", "score": 6, "verdict_ko": "오버더숄더 구도는 적절하나 전경 좌측에 식별하기 어려운 회색 물체가 시야를 일부 가립니다." }, { "label": "D", "score": 5, "verdict_ko": "레퍼런스의 의상(비니)을 잘 재현했으나, 앵글이 다소 넓어 몰입감 있는 시점 쇼트로 보기 어렵습니다." }, { "label": "B", "score": 4, "verdict_ko": "카메라 구도가 미디엄 쇼트에 가까워 요구된 오버더숄더 시점 쇼트의 조건을 충족하지 못합니다." } ] }, "combined": { "totals": { "A": 14, "B": 10, "C": 9, "D": 9 }, "ranking": [ "A", "B", "C", "D" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 10, "C": 9, "D": 9 }, "selected": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "어깨를 전경에 크게 걸쳐 우측 벽면의 원형 표식들을 강조하는 '어깨 너머 시점 쇼트'의 카메라 구도를 가장 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "의상과 로케이션 일치도는 매우 높으나, 인물이 프레임 안쪽으로 들어가 있어 지정된 오버더숄더 구도보다는 일반적인 후면 쇼트에 가깝습니다." }, { "label": "C", "score": 3, "verdict_ko": "레퍼런스의 조타실 콘솔 디자인과 우측 벽면/출입구 구조를 임의로 왜곡하여 로케이션 고정 지시를 위반했습니다." }, { "label": "D", "score": 4, "verdict_ko": "인물이 조타석에 앉아 정면을 향하는 넓은 후면 쇼트로 연출되어, 벽면 표식을 향한 어깨 너머 시점 쇼트라는 핵심 구도를 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/61621f50-9d47-4fdc-9076-02a839c1733c/episodes/a661d23e-a05d-452e-9a28-1153e69f01fd/images/background_chain/L19B01.png" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [] }, "fix_skipped": true, "variant_map": { "A": { "variant": 0, "conti": true }, "B": { "variant": 1, "conti": true }, "C": { "variant": 0, "conti": false }, "D": { "variant": 1, "conti": false } }, "conti_winner": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·A=변형0)" }, "S28sh8::variants": { "author_fp": "8a08f438f3e4bdc3", "author": { "variants": [ { "approach_ko": "인우의 굳은 눈과 어깨에 맞닿은 수리영의 머리를 압축한 정면 밀착 클로즈업.", "prompt_en": "Photograph this as a tight, nearly frontal close-up at Inwoo’s eye level, cropping closely around his face, shoulder, and Suri-young’s resting head. Use a restrained portrait-lens feel with shallow depth, holding Inwoo’s rigid, wide-eyed expression in crisp focus while the point of contact remains equally legible. Let soft dusk illumination from the wheelhouse windows model their faces with subdued contrast, while the circular-marked interior recedes into an unobtrusive blur." }, { "approach_ko": "두 사람의 정지된 접촉을 측면에서 포착하고 좁은 조타실의 깊이와 황혼 바다를 함께 살리는 구도.", "prompt_en": "Take a side-on medium shot from slightly below shoulder height, using layered depth to place the two figures against the wheelhouse windows and circular-marked surfaces. Keep their shared silhouette and the downward pull of Suri-young’s slack body visually clear, while Inwoo’s fixed gaze cuts across the frame rather than toward the camera. Use a broader lens feel and deeper focus so the cramped interior geometry and dusk beyond the windows intensify the suspended stillness." } ] }, "reused": false }, "S28sh8": { "input_fingerprint": "81fdf106057271c8", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 눈을 부릅뜨고 굳어 있는 인우의 어깨 위로 수리영의 머리가 맞닿아 있는 찰나의 정지 구도.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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 wheelhouse materials, dusk light, wall covered with circular markings, and the man's unchanged clothing and position. Preserve the woman's clothing and her collapse against his shoulder. Exclude the old photograph, the pointing hand, and the earlier over-the-shoulder viewing action.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Her body has gone slack as she collapses against Inwoo, with her head resting on his shoulder and her torso and limbs sagging downward.\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): Suri-young is unconscious; her body has gone slack as she collapses against Inwoo, with her head resting on his shoulder and her torso and limbs sagging downward; no dropping of her photograph, backpack, or maps is shown, so those belongings remain with her 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 눈을 부릅뜨고 굳어 있는 인우의 어깨 위로 수리영의 머리가 맞닿아 있는 찰나의 정지 구도.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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 wheelhouse materials, dusk light, wall covered with circular markings, and the man's unchanged clothing and position. Preserve the woman's clothing and her collapse against his shoulder. Exclude the old photograph, the pointing hand, and the earlier over-the-shoulder viewing action.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Her body has gone slack as she collapses against Inwoo, with her head resting on his shoulder and her torso and limbs sagging downward.\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): Suri-young is unconscious; her body has gone slack as she collapses against Inwoo, with her head resting on his shoulder and her torso and limbs sagging downward; no dropping of her photograph, backpack, or maps is shown, so those belongings remain with her 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph this as a tight, nearly frontal close-up at Inwoo’s eye level, cropping closely around his face, shoulder, and Suri-young’s resting head. Use a restrained portrait-lens feel with shallow depth, holding Inwoo’s rigid, wide-eyed expression in crisp focus while the point of contact remains equally legible. Let soft dusk illumination from the wheelhouse windows model their faces with subdued contrast, while the circular-marked interior recedes into an unobtrusive blur.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 눈을 부릅뜨고 굳어 있는 인우의 어깨 위로 수리영의 머리가 맞닿아 있는 찰나의 정지 구도.\n\nLOCATION (lock): A cramped fishing-boat wheelhouse containing steering controls and windows overlooking the sea. Numerous circular markings cover the interior surfaces. 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 wheelhouse materials, dusk light, wall covered with circular markings, and the man's unchanged clothing and position. Preserve the woman's clothing and her collapse against his shoulder. Exclude the old photograph, the pointing hand, and the earlier over-the-shoulder viewing action.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Her body has gone slack as she collapses against Inwoo, with her head resting on his shoulder and her torso and limbs sagging downward.\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): Suri-young is unconscious; her body has gone slack as she collapses against Inwoo, with her head resting on his shoulder and her torso and limbs sagging downward; no dropping of her photograph, backpack, or maps is shown, so those belongings remain with her 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: 수리영 (한국인 여성, 21세, 짙은색 머리, 젊은 얼굴); 인우 (한국인 남성, 24세, 짙은색 머리, 단정한 미청년 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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTake a side-on medium shot from slightly below shoulder height, using layered depth to place the two figures against the wheelhouse windows and circular-marked surfaces. Keep their shared silhouette and the downward pull of Suri-young’s slack body visually clear, while Inwoo’s fixed gaze cuts across the frame rather than toward the camera. Use a broader lens feel and deeper focus so the cramped interior geometry and dusk beyond the windows intensify the suspended stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시문의 핵심인 눈을 부릅뜬 인우의 표정과 머리가 맞닿은 근접 구도를 잘 구현했으나 수리영의 헬멧이 누락됨." }, { "label": "B", "score": 4, "verdict_ko": "요구된 밀착된 정지 구도와 표정 대신 불필요하게 넓은 앵글을 잡았으며 인우의 비니가 누락됨." } ], "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/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/scene/recipe/S28sh5_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영의 캐릭터 레퍼런스에 포함된 검은색 자전거 헬멧이 누락되었습니다.", "fix_en": "Add the black bicycle helmet to Suri-young's head, exactly matching her character reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the black bicycle helmet to Suri-young's head, exactly matching her character reference.\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.", "ref_mode": "prev+엔티티" }, "S29sh3::variants": { "author_fp": "291b887c2cbf8c15", "author": { "variants": [ { "approach_ko": "수면 높이의 밀착된 측면 촬영으로 뱃머리와 안개가 맞부딪히는 물리적 순간을 강조한다.", "prompt_en": "Photograph from a low, waterline-height lateral position in a tight side-profile frame, making the bow’s contact with the fog bank the dominant visual event. Use an intimate, moderately wide lens feel with strong foreground water texture and layered depth through the hull and fog; let diffused golden sunset light rake softly across the visible surfaces while the disturbed water remains crisp and tactile." }, { "approach_ko": "멀리 떨어진 정적인 측면 구도로 배와 거대한 안개 경계의 압도적인 규모 차이를 보여준다.", "prompt_en": "Use a distant, static broadside composition with the boat held relatively small against a large field of sea and dense fog, emphasizing the vessel’s passage across the stark atmospheric boundary. Favor a compressed lens feel and restrained depth, with the golden sunset reduced to a muted glow inside the fog; preserve clean lateral geometry and allow the bow-driven water disturbance to provide the frame’s concentrated point of motion." } ] }, "reused": false }, "S29sh3": { "input_fingerprint": "581e269ac83b0dd6", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): sunset, golden light and dense fog.\n\nSHOT TEXT (authoritative, Korean): 배의 앞머리 절반이 짙은 안개 속에 파묻힌 채, 뱃머리에서 물살을 일으키며 안개 속으로 진입하는 순간의 측면 컷.\n\nLOCATION (lock): Open sea crossed by a small fishing boat, with a dense offshore fog bank forming a visual boundary across the water. 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): Suri-young remains unconscious aboard In-woo’s boat with her backpack, maps, and old photograph still with her as the vessel enters the dense fog.\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.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): sunset, golden light and dense fog.\n\nSHOT TEXT (authoritative, Korean): 배의 앞머리 절반이 짙은 안개 속에 파묻힌 채, 뱃머리에서 물살을 일으키며 안개 속으로 진입하는 순간의 측면 컷.\n\nLOCATION (lock): Open sea crossed by a small fishing boat, with a dense offshore fog bank forming a visual boundary across the water. 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): Suri-young remains unconscious aboard In-woo’s boat with her backpack, maps, and old photograph still with her as the vessel enters the dense fog.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a low, waterline-height lateral position in a tight side-profile frame, making the bow’s contact with the fog bank the dominant visual event. Use an intimate, moderately wide lens feel with strong foreground water texture and layered depth through the hull and fog; let diffused golden sunset light rake softly across the visible surfaces while the disturbed water remains crisp and tactile.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — coastal South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): sunset, golden light and dense fog.\n\nSHOT TEXT (authoritative, Korean): 배의 앞머리 절반이 짙은 안개 속에 파묻힌 채, 뱃머리에서 물살을 일으키며 안개 속으로 진입하는 순간의 측면 컷.\n\nLOCATION (lock): Open sea crossed by a small fishing boat, with a dense offshore fog bank forming a visual boundary across the water. 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): Suri-young remains unconscious aboard In-woo’s boat with her backpack, maps, and old photograph still with her as the vessel enters the dense fog.\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.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a distant, static broadside composition with the boat held relatively small against a large field of sea and dense fog, emphasizing the vessel’s passage across the stark atmospheric boundary. Favor a compressed lens feel and restrained depth, with the golden sunset reduced to a muted glow inside the fog; preserve clean lateral geometry and allow the bow-driven water disturbance to provide the frame’s concentrated point of motion.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 8, "B": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "스토리보드가 제시한 정확한 측면 컷 구도를 유지하며, 뱃머리 절반이 안개에 파묻히는 묘사를 충실하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍이 과도하게 넓어져 스토리보드의 구도를 벗어났으며, 뱃머리가 안개 벽에 진입하는 느낌이 부족합니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/61621f50-9d47-4fdc-9076-02a839c1733c/images/a661d23e-a05d-452e-9a28-1153e69f01fd/conti/lane_sketch_S29sh3.png" } ], "critique": { "issues": [ { "issue_ko": "뱃머리 절반이 안개에 파묻혀야 한다는 지시와 달리, 뱃머리 전체가 안개 밖으로 완전히 드러나 있음.", "fix_en": "Extend the dense white fog bank to the left so that it completely covers and obscures the front half of the boat's bow." }, { "issue_ko": "인물이 등장하지 않아야 하는 컷임에도 갑판 위 마스트 근처에 사람의 실루엣이 서 있음.", "fix_en": "Remove the silhouette of the person standing on the deck near the mast structure." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Extend the dense white fog bank to the left so that it completely covers and obscures the front half of the boat's bow.\n- Remove the silhouette of the person standing on the deck near the mast structure.\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.", "ref_mode": "lane(map_marker): 스케치" } }