{ "S1sh3::variants": { "author_fp": "ef3bfdbd2b593778", "author": { "variants": [ { "approach_ko": "눈높이의 극단적 정면 클로즈업으로 얼굴과 공포 반응을 대칭적으로 압박한다.", "prompt_en": "Use an eye-level, extreme frontal close-up, with the face nearly filling the frame and the eyes held close to the upper third. Keep the composition rigidly centered and confrontational, with very shallow focus dissolving the forest into dim dusk texture. Soft, cool ambient light falls evenly enough to preserve natural detail in both eyes while gentle undergrowth shadow shapes the face." }, { "approach_ko": "가까운 광각 정면 숏으로 얼굴을 전경에 두고 뒤의 탈출 경로를 깊게 드러낸다.", "prompt_en": "Photograph from close range with a moderately wide perspective, keeping the face dominant and square to camera while allowing the head, shoulders, and receding escape path to remain legible. Build strong depth with nearby vegetation framing the foreground and the forest drawing away directly behind him. Let the fading dusk create uneven, natural patches of cool light and shadow, making the surrounding space feel immediate and inescapable." } ] }, "reused": false }, "S1sh3": { "input_fingerprint": "c20e5ce2fe958cc1", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜬 채 겁에 질린 숲속 사건의 남자 피해자의 얼굴 정면 구도\n\nLOCATION (lock): A dense forest with thick undergrowth forming a rough escape path through the vegetation. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 male victim is missing one shoe and has a large, bleeding bite mark on his thigh.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속 사건의 남자 피해자 (한국인 남성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜬 채 겁에 질린 숲속 사건의 남자 피해자의 얼굴 정면 구도\n\nLOCATION (lock): A dense forest with thick undergrowth forming a rough escape path through the vegetation. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 male victim is missing one shoe and has a large, bleeding bite mark on his thigh.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속 사건의 남자 피해자 (한국인 남성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an eye-level, extreme frontal close-up, with the face nearly filling the frame and the eyes held close to the upper third. Keep the composition rigidly centered and confrontational, with very shallow focus dissolving the forest into dim dusk texture. Soft, cool ambient light falls evenly enough to preserve natural detail in both eyes while gentle undergrowth shadow shapes 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜬 채 겁에 질린 숲속 사건의 남자 피해자의 얼굴 정면 구도\n\nLOCATION (lock): A dense forest with thick undergrowth forming a rough escape path through the vegetation. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 male victim is missing one shoe and has a large, bleeding bite mark on his thigh.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 숲속 사건의 남자 피해자 (한국인 남성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from close range with a moderately wide perspective, keeping the face dominant and square to camera while allowing the head, shoulders, and receding escape path to remain legible. Build strong depth with nearby vegetation framing the foreground and the forest drawing away directly behind him. Let the fading dusk create uneven, natural patches of cool light and shadow, making the surrounding space feel immediate and inescapable.\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": "요청된 완벽한 정면 얼굴 구도와 캐릭터의 외형(비니 등)을 레퍼런스와 정확히 일치시켰습니다." }, { "label": "B", "score": 4, "verdict_ko": "지정된 정면 얼굴 구도보다 더 넓은 프레임을 사용했으며, 레퍼런스의 핵심인 비니를 누락하여 캐릭터 일관성을 어겼습니다." } ], "refs": [ { "label": "CHARACTER REFERENCE — 숲속 사건의 남자 피해자: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [] }, "fix_skipped": true, "ref_mode": "플레이트+콘티+엔티티", "share_plan": { "ref_plan": "background" } }, "S1sh4::variants": { "author_fp": "0efbf38e0c05a38f", "author": { "variants": [ { "approach_ko": "피해자 곁의 낮은 시점에서 위로 올려다보며 공중의 습격자가 덮쳐오는 위압감을 강조한다.", "prompt_en": "Shoot from an extremely low position beside the fallen victim, looking upward through the rough escape path so the victim anchors the near foreground while the airborne attacker dominates above, both outstretched arms driving toward him. Use a wide, immersive lens feel with strong near-to-far depth; preserve readable human anatomy and clothing in the attacker while the dusk backlight keeps his face difficult to identify. Let the dense undergrowth close around the composition, making the suspended instant feel immediate and inescapable." }, { "approach_ko": "숲길 옆의 먼 측면 시점과 압축된 원근으로 공중의 습격자와 쓰러진 피해자의 간격을 냉정하게 포착한다.", "prompt_en": "Frame the moment from a distant lateral viewpoint through the forest, using a compressed lens feel to hold the airborne attacker and fallen victim in a stark vertical relationship within the same shallow plane. Keep both bodies clearly legible in profile, emphasizing the attacker’s extended arms and the narrowing gap toward the victim rather than subjective impact. Dusk light filters softly across the vegetation while foreground foliage partially hems the frame, creating an observational, ominously restrained image." } ] }, "reused": false }, "S1sh4": { "input_fingerprint": "92cd6fc319eb65b4", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 공중에 뜬 검은 실루엣의 정체불명의 습격자가 바닥에 쓰러진 숲속 사건의 남자 피해자를 향해 양팔을 뻗은 찰나\n\nLOCATION (lock): A dense forest with thick undergrowth forming a rough escape path through the vegetation. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fallen male victim must still be missing one shoe, with the large bite wound on his thigh continuing to bleed.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 습격자 (한국인으로 보이는 성인 남성, 어둠에 가려진 얼굴, 식별하기 어려운 머리 형태, 붉은 눈동자를 제외한 이목구비 불명); 숲속 사건의 남자 피해자 (한국인 남성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 공중에 뜬 검은 실루엣의 정체불명의 습격자가 바닥에 쓰러진 숲속 사건의 남자 피해자를 향해 양팔을 뻗은 찰나\n\nLOCATION (lock): A dense forest with thick undergrowth forming a rough escape path through the vegetation. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fallen male victim must still be missing one shoe, with the large bite wound on his thigh continuing to bleed.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 습격자 (한국인으로 보이는 성인 남성, 어둠에 가려진 얼굴, 식별하기 어려운 머리 형태, 붉은 눈동자를 제외한 이목구비 불명); 숲속 사건의 남자 피해자 (한국인 남성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from an extremely low position beside the fallen victim, looking upward through the rough escape path so the victim anchors the near foreground while the airborne attacker dominates above, both outstretched arms driving toward him. Use a wide, immersive lens feel with strong near-to-far depth; preserve readable human anatomy and clothing in the attacker while the dusk backlight keeps his face difficult to identify. Let the dense undergrowth close around the composition, making the suspended instant feel immediate and inescapable.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 공중에 뜬 검은 실루엣의 정체불명의 습격자가 바닥에 쓰러진 숲속 사건의 남자 피해자를 향해 양팔을 뻗은 찰나\n\nLOCATION (lock): A dense forest with thick undergrowth forming a rough escape path through the vegetation. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The fallen male victim must still be missing one shoe, with the large bite wound on his thigh continuing to bleed.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 습격자 (한국인으로 보이는 성인 남성, 어둠에 가려진 얼굴, 식별하기 어려운 머리 형태, 붉은 눈동자를 제외한 이목구비 불명); 숲속 사건의 남자 피해자 (한국인 남성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame the moment from a distant lateral viewpoint through the forest, using a compressed lens feel to hold the airborne attacker and fallen victim in a stark vertical relationship within the same shallow plane. Keep both bodies clearly legible in profile, emphasizing the attacker’s extended arms and the narrowing gap toward the victim rather than subjective impact. Dusk light filters softly across the vegetation while foreground foliage partially hems the frame, creating an observational, ominously restrained image.\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": 4 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "스토리보드의 수평 공중 부양 자세와 어두운 실루엣, 피해자가 신발을 한 짝만 잃어버린 상태를 모두 정확히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "피해자의 신발이 두 짝 모두 벗겨져 있고 습격자가 실루엣으로 묘사되지 않아 프롬프트 지시를 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/lane_sketch_S1sh4.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S1sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 정체불명의 습격자: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 숲속 사건의 남자 피해자: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "공중에 뜬 습격자의 얼굴이 밝게 노출되어 이목구비와 머리 형태가 뚜렷하게 보이며, 붉은 눈동자 설정이 반영되지 않았습니다.", "fix_en": "Darken the flying attacker's entire head and face into a heavy, featureless black shadow, adding only distinct red pupils to his eyes." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Darken the flying attacker's entire head and face into a heavy, featureless black shadow, adding only distinct red pupils to his eyes.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S1sh3" } }, "S2sh1::variants": { "author_fp": "0b4f45eba925f604", "author": { "variants": [ { "approach_ko": "전봇대 가까이에서 비스듬히 파고드는 밀착 촬영으로 두 인물의 얽힘과 피해자의 압박감을 강조한다.", "prompt_en": "Photograph the struggle from close range at roughly shoulder height, looking obliquely along the utility pole so it becomes a hard vertical anchor in the composition. Use a tight, claustrophobic crop and a moderately shallow depth of field, with the attacker’s near-side body partially dominating the foreground while the victim’s face, hair, and pinned posture remain clearly readable. Let the dim streetlight rake unevenly across the figures, preserving deep facial obscurity on the attacker and concentrating visual tension on their compressed physical entanglement." }, { "approach_ko": "골목 깊숙이 떨어진 넓은 시점에서 어둠과 빈 공간 속 전봇대의 습격 장면을 고립된 사건처럼 포착한다.", "prompt_en": "Frame the moment from farther down the alley in a wide, static composition, keeping the utility pole and the entangled pair as a stark focal cluster within the surrounding depth and empty alley space. Use a slightly elevated eye line and deep focus so the narrow passage, repeating poles, and sparse streetlights establish distance and isolation without diminishing the legibility of the pinning action. Hold most of the frame in restrained darkness, allowing the limited streetlight to isolate the figures with a cold, observational sense of danger." } ] }, "reused": false }, "S2sh1": { "input_fingerprint": "ad6bdb307837409c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim streetlights.\n\nSHOT TEXT (authoritative, Korean): 희미한 가로등 불빛 아래, 정체불명의 습격자가 뒷골목 사건의 여자 피해자를 전봇대에 거칠게 밀어붙인 채 뒤엉켜 있는 구도\n\nLOCATION (lock): A narrow urban back alley lined with utility poles, including one where two figures are entangled. Sparse streetlights provide the only stated exterior illumination. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 remains pinned against the utility pole and entangled with her attacker.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 습격자 (한국인으로 보이는 성인 남성, 어둠에 가려진 얼굴, 식별하기 어려운 머리 형태, 붉은 눈동자를 제외한 이목구비 불명); 뒷골목 사건의 여자 피해자 (한국인 여성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim streetlights.\n\nSHOT TEXT (authoritative, Korean): 희미한 가로등 불빛 아래, 정체불명의 습격자가 뒷골목 사건의 여자 피해자를 전봇대에 거칠게 밀어붙인 채 뒤엉켜 있는 구도\n\nLOCATION (lock): A narrow urban back alley lined with utility poles, including one where two figures are entangled. Sparse streetlights provide the only stated exterior illumination. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 remains pinned against the utility pole and entangled with her attacker.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 습격자 (한국인으로 보이는 성인 남성, 어둠에 가려진 얼굴, 식별하기 어려운 머리 형태, 붉은 눈동자를 제외한 이목구비 불명); 뒷골목 사건의 여자 피해자 (한국인 여성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the struggle from close range at roughly shoulder height, looking obliquely along the utility pole so it becomes a hard vertical anchor in the composition. Use a tight, claustrophobic crop and a moderately shallow depth of field, with the attacker’s near-side body partially dominating the foreground while the victim’s face, hair, and pinned posture remain clearly readable. Let the dim streetlight rake unevenly across the figures, preserving deep facial obscurity on the attacker and concentrating visual tension on their compressed physical entanglement.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim streetlights.\n\nSHOT TEXT (authoritative, Korean): 희미한 가로등 불빛 아래, 정체불명의 습격자가 뒷골목 사건의 여자 피해자를 전봇대에 거칠게 밀어붙인 채 뒤엉켜 있는 구도\n\nLOCATION (lock): A narrow urban back alley lined with utility poles, including one where two figures are entangled. Sparse streetlights provide the only stated exterior illumination. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 remains pinned against the utility pole and entangled with her attacker.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 습격자 (한국인으로 보이는 성인 남성, 어둠에 가려진 얼굴, 식별하기 어려운 머리 형태, 붉은 눈동자를 제외한 이목구비 불명); 뒷골목 사건의 여자 피해자 (한국인 여성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame the moment from farther down the alley in a wide, static composition, keeping the utility pole and the entangled pair as a stark focal cluster within the surrounding depth and empty alley space. Use a slightly elevated eye line and deep focus so the narrow passage, repeating poles, and sparse streetlights establish distance and isolation without diminishing the legibility of the pinning action. Hold most of the frame in restrained darkness, allowing the limited streetlight to isolate the figures with a cold, observational sense of danger.\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": "'뒤엉켜 있는 구도'를 밀착된 앵글로 긴장감 있게 살렸으며, 습격자의 얼굴을 어둠 속에 두어 프롬프트 지시를 비교적 잘 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "와이드 샷으로 인해 피사체 간의 뒤엉킨 느낌이 약화되었으며, 습격자의 얼굴이 가로등에 너무 밝게 드러나 '어둠에 가려진 얼굴' 지시를 위반했습니다." } ], "refs": [ { "label": "CHARACTER REFERENCE — 정체불명의 습격자: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 뒷골목 사건의 여자 피해자: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "화면 오른쪽 전봇대를 잡고 있는 세 번째 손(검은색 소매)이 생성되어 해부학적으로 심각한 오류가 발생했습니다.", "fix_en": "Remove the anatomically impossible extra hand and black sleeve from the right side of the utility pole." }, { "issue_ko": "전봇대에 부착된 스티커 및 종이에 문자가 포함되어 있어 모든 형태의 텍스트를 금지한 규칙을 위반했습니다.", "fix_en": "Remove all visible text, letters, and characters from the utility pole." }, { "issue_ko": "지시문에 명시된 습격자의 '붉은 눈동자'가 묘사되지 않았습니다.", "fix_en": "Change the attacker's eye in the shadow so the iris is visibly red, maintaining normal anatomical proportions." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the anatomically impossible extra hand and black sleeve from the right side of the utility pole.\n- Remove all visible text, letters, and characters from the utility pole.\n- Change the attacker's eye in the shadow so the iris is visibly red, maintaining normal anatomical proportions.\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", "share_plan": { "ref_plan": "background" } }, "S2sh3::variants": { "author_fp": "8d66dae3494a3c4e", "author": { "variants": [ { "approach_ko": "목의 물림 지점과 두 인물의 측면 윤곽에 집중하는 숨 막히는 초근접 촬영.", "prompt_en": "Photograph the instant in an unflinching, eye-level extreme close side view, cropping tightly around the two profiles, jaws, neck, and the victim’s contact with the pole. Place the bite at the sharp compositional center, with very shallow depth of field letting the attacker’s obscured face and alley textures fall away rapidly. Shape the sparse streetlight as a narrow grazing edge across skin, teeth, and the torn wound, preserving heavy natural shadow and a tense, tactile realism." }, { "approach_ko": "낮은 측면 중경에서 전봇대와 골목의 깊이를 함께 살려 도주 불가능한 상황을 강조하는 촬영.", "prompt_en": "Use a low, restrained medium side shot that keeps the entangled upper bodies and utility pole clearly legible while the narrow alley recedes behind them. Compose the pole as a rigid vertical that traps the victim in frame, with the attacker’s body forming a dark diagonal into the bite; hold greater depth of field so the worn alley surfaces and sparse streetlights remain present without competing with the action. Let the dim streetlight separate their edges from the background while the faces remain largely submerged in credible night shadow." } ] }, "reused": false }, "S2sh3": { "input_fingerprint": "833ddf99ebc5814e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim streetlights.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 습격자의 이빨이 뒷골목 사건의 여자 피해자의 목에 깊이 박혀 살점을 강하게 물어뜯는 찰나의 측면\n\nLOCATION (lock): A narrow urban back alley lined with utility poles, including one where two figures are entangled. Sparse streetlights provide the only stated exterior illumination. 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 dim streetlamp lighting, utility pole, alley textures, and the attackers' continuing clothing and positions from the reference. Exclude the earlier wide pushing pose and reframe the action tightly around the bite at the woman's neck.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 remains pressed against the utility pole as the attacker bites deeply into her neck, leaving a torn neck wound.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 습격자 (한국인으로 보이는 성인 남성, 어둠에 가려진 얼굴, 식별하기 어려운 머리 형태, 붉은 눈동자를 제외한 이목구비 불명); 뒷골목 사건의 여자 피해자 (한국인 여성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim streetlights.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 습격자의 이빨이 뒷골목 사건의 여자 피해자의 목에 깊이 박혀 살점을 강하게 물어뜯는 찰나의 측면\n\nLOCATION (lock): A narrow urban back alley lined with utility poles, including one where two figures are entangled. Sparse streetlights provide the only stated exterior illumination. 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 dim streetlamp lighting, utility pole, alley textures, and the attackers' continuing clothing and positions from the reference. Exclude the earlier wide pushing pose and reframe the action tightly around the bite at the woman's neck.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 remains pressed against the utility pole as the attacker bites deeply into her neck, leaving a torn neck wound.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 습격자 (한국인으로 보이는 성인 남성, 어둠에 가려진 얼굴, 식별하기 어려운 머리 형태, 붉은 눈동자를 제외한 이목구비 불명); 뒷골목 사건의 여자 피해자 (한국인 여성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the instant in an unflinching, eye-level extreme close side view, cropping tightly around the two profiles, jaws, neck, and the victim’s contact with the pole. Place the bite at the sharp compositional center, with very shallow depth of field letting the attacker’s obscured face and alley textures fall away rapidly. Shape the sparse streetlight as a narrow grazing edge across skin, teeth, and the torn wound, preserving heavy natural shadow and a tense, tactile 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.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim streetlights.\n\nSHOT TEXT (authoritative, Korean): 정체불명의 습격자의 이빨이 뒷골목 사건의 여자 피해자의 목에 깊이 박혀 살점을 강하게 물어뜯는 찰나의 측면\n\nLOCATION (lock): A narrow urban back alley lined with utility poles, including one where two figures are entangled. Sparse streetlights provide the only stated exterior illumination. 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 dim streetlamp lighting, utility pole, alley textures, and the attackers' continuing clothing and positions from the reference. Exclude the earlier wide pushing pose and reframe the action tightly around the bite at the woman's neck.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 remains pressed against the utility pole as the attacker bites deeply into her neck, leaving a torn neck wound.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 정체불명의 습격자 (한국인으로 보이는 성인 남성, 어둠에 가려진 얼굴, 식별하기 어려운 머리 형태, 붉은 눈동자를 제외한 이목구비 불명); 뒷골목 사건의 여자 피해자 (한국인 여성, 성인, 평범한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a low, restrained medium side shot that keeps the entangled upper bodies and utility pole clearly legible while the narrow alley recedes behind them. Compose the pole as a rigid vertical that traps the victim in frame, with the attacker’s body forming a dark diagonal into the bite; hold greater depth of field so the worn alley surfaces and sparse streetlights remain present without competing with the action. Let the dim streetlight separate their edges from the background while the faces remain largely submerged in credible night 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": 3, "B": 8 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "남성의 머리 윗부분이 잘려 나간 치명적인 이미지 오류가 있으며, 목을 무는 행동이 전혀 연출되지 않았고 붉은 눈이 잘못된 인물에게 적용되었습니다." }, { "label": "B", "score": 8, "verdict_ko": "지정된 골목길 배경과 전봇대 위치를 잘 유지하며, 습격자가 피해자의 목을 무는 결정적인 순간과 붉은 눈의 디테일을 프롬프트에 맞게 훌륭하게 구현했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S2sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 정체불명의 습격자: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 뒷골목 사건의 여자 피해자: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "습격자의 얼굴과 머리 형태가 어둠에 가려지지 않고 이목구비가 선명하게 드러나 있습니다.", "fix_en": "Obscure the attacker's face and head completely in deep shadow so only the red eye remains identifiable." }, { "issue_ko": "습격자의 이빨이 피해자의 목에 깊이 박혀 살점을 물어뜯지 않고, 턱 아래에 입만 벌린 채 떠 있습니다.", "fix_en": "Embed the attacker's teeth deeply into the flesh of the victim's neck to show a forceful bite." }, { "issue_ko": "피해자의 목을 쥐고 있는 습격자의 손이 해부학적으로 어긋나며(오른팔 위치에 왼손 형태), 손가락 형태가 심하게 왜곡되어 있습니다.", "fix_en": "Redraw the attacker's hand on the neck to be an anatomically correct right hand gripping the neck." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Obscure the attacker's face and head completely in deep shadow so only the red eye remains identifiable.\n- Embed the attacker's teeth deeply into the flesh of the victim's neck to show a forceful bite.\n- Redraw the attacker's hand on the neck to be an anatomically correct right hand gripping the neck.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S2sh1" } }, "S3sh2::variants": { "author_fp": "0c2104bf2f135b5d", "author": { "variants": [ { "approach_ko": "들것과 덮개 천을 중심에 둔 수직 톱숏으로 시신의 고요함을 냉정하게 강조한다.", "prompt_en": "Photograph the stretcher from a close, nearly vertical overhead position, with the cloth-covered corpse dominating the frame. Use a restrained, geometrically ordered composition and crisp natural daylight, holding enough surrounding ground at the edges to anchor the physical crime-scene location while emphasizing absolute stillness." }, { "approach_ko": "높은 사선 시점의 넓은 구도로 들것과 주변 범죄 현장 사이의 고립감을 부각한다.", "prompt_en": "Use a wider high-oblique view that looks down along the length of the stretcher, placing the covered corpse off center within the visible crime-scene surroundings. Build depth through the diagonal stretcher and receding ground, with subdued daylight contrast and selective focus that isolates the covered form without obscuring the established location." } ] }, "reused": false }, "S3sh2": { "input_fingerprint": "7f31fe6b7aa201b3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 들것 위에 천으로 덮인 시신을 내려다보는 구도\n\nLOCATION (lock): Two outdoor crime scenes—a forest trail area and an urban back alley—sealed with police lines as officers examine the ground and recover bodies. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 corpse remains motionless on the stretcher beneath its covering cloth as it is collected.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 들것 위에 천으로 덮인 시신을 내려다보는 구도\n\nLOCATION (lock): Two outdoor crime scenes—a forest trail area and an urban back alley—sealed with police lines as officers examine the ground and recover bodies. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 corpse remains motionless on the stretcher beneath its covering cloth as it is collected.\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 stretcher from a close, nearly vertical overhead position, with the cloth-covered corpse dominating the frame. Use a restrained, geometrically ordered composition and crisp natural daylight, holding enough surrounding ground at the edges to anchor the physical crime-scene location while emphasizing absolute 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 들것 위에 천으로 덮인 시신을 내려다보는 구도\n\nLOCATION (lock): Two outdoor crime scenes—a forest trail area and an urban back alley—sealed with police lines as officers examine the ground and recover bodies. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 corpse remains motionless on the stretcher beneath its covering cloth as it is collected.\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 high-oblique view that looks down along the length of the stretcher, placing the covered corpse off center within the visible crime-scene surroundings. Build depth through the diagonal stretcher and receding ground, with subdued daylight contrast and selective focus that isolates the covered form without obscuring the established location.\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": "화면에 산 사람이 없어야 한다는 지침과 들것을 내려다보는 하이 앵글 구도를 정확히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "지침에 어긋나게 배경에 경찰관들이 등장하며, 구도 역시 내려다보는 각도가 부족하고 폴리스 라인에 깨진 문자가 포함되었습니다." } ], "refs": [], "critique": { "issues": [ { "issue_ko": "카메라 앵글이 거의 수직으로 내려다보는 오버헤드(부감) 구도가 아닙니다.", "fix_en": "Change the camera angle to a nearly vertical overhead position looking directly down at the stretcher." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the camera angle to a nearly vertical overhead position looking directly down at the stretcher.\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)", "share_plan": { "ref_plan": "background" } }, "S4sh1::variants": { "author_fp": "bbf28b3ce9bcb1d9", "author": { "variants": [ { "approach_ko": "가파른 수직 부감의 넓은 프레임으로 옥탑방과 옥상 구조의 기하학적 관계를 강조한다.", "prompt_en": "Photograph from a steep, near-vertical overhead position in a wide establishing frame. Let the rooftop dwelling anchor the composition while the surrounding roof plane and building edges form strong geometric layers, with crisp spatial separation and restrained documentary realism. Use the existing daylight to reveal surface texture without dramatizing or altering the location." }, { "approach_ko": "멀리 떨어진 사선 부감과 압축된 원근으로 옥탑방이 주변 환경에 눌린 듯 보이게 담는다.", "prompt_en": "Use a distant, oblique high-angle viewpoint with compressed perspective, framing the rooftop dwelling within the layered surroundings rather than looking straight down. Keep the composition dense and observational, allowing foreground and background planes to visually press around the small structure. Preserve the natural direction, contrast, and color of the established daylight for a quiet, lived-in urban mood." } ] }, "reused": false }, "S4sh1": { "input_fingerprint": "d1d32ccd0539e22d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 도심 변두리의 오래된 빌라 옥탑방이 내려다보이는 부감 구도\n\nLOCATION (lock): The rooftop level of an old low-rise residential building on the outskirts of a city, with a small rooftop dwelling visible in an elevated establishing view. 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 도심 변두리의 오래된 빌라 옥탑방이 내려다보이는 부감 구도\n\nLOCATION (lock): The rooftop level of an old low-rise residential building on the outskirts of a city, with a small rooftop dwelling visible in an elevated establishing view. 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 from a steep, near-vertical overhead position in a wide establishing frame. Let the rooftop dwelling anchor the composition while the surrounding roof plane and building edges form strong geometric layers, with crisp spatial separation and restrained documentary realism. Use the existing daylight to reveal surface texture without dramatizing or altering the location.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 도심 변두리의 오래된 빌라 옥탑방이 내려다보이는 부감 구도\n\nLOCATION (lock): The rooftop level of an old low-rise residential building on the outskirts of a city, with a small rooftop dwelling visible in an elevated establishing view. 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 distant, oblique high-angle viewpoint with compressed perspective, framing the rooftop dwelling within the layered surroundings rather than looking straight down. Keep the composition dense and observational, allowing foreground and background planes to visually press around the small structure. Preserve the natural direction, contrast, and color of the established daylight for a quiet, lived-in urban mood.\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": 7 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "지정된 '건물 외관(Structure Look)' 레퍼런스를 따르지 않고 '장소(Location)' 사진의 옥탑 구조와 물탱크를 잘못 복제했습니다." }, { "label": "B", "score": 7, "verdict_ko": "요청된 부감 구도를 잘 구현했으며, 지침에 따라 '건물 외관' 레퍼런스의 옥탑방 형태(녹색 문, 작은 창문, 외부 계단 등)를 정확하게 반영했습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L04B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/background_chain/seed_bg_rooftop_villa_complex_sel.png" } ], "critique": { "issues": [ { "issue_ko": "메인 건물 주변의 골목길, 인접 건물, 배경 환경이 LOCATION 사진이 아닌 STRUCTURE LOOK 사진의 주변 환경을 잘못 복사했습니다.", "fix_en": "Replace all surrounding streets, adjacent buildings, and the background environment to exactly match the surroundings shown in the LOCATION photograph." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace all surrounding streets, adjacent buildings, and the background environment to exactly match the surroundings shown in the LOCATION 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)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02" }, "assigned": "L04B01", "choice": "A", "confident": true, "reason_ko": "원문에서 '내려다보이는 부감 구도'를 요구하고 있으므로, 옥탑방 전체를 위에서 내려다보는 앵글인 A가 적합합니다.", "chosen": "L04B01" } }, "S5sh9::variants": { "author_fp": "863134a313453238", "author": { "variants": [ { "approach_ko": "눈높이의 밀착 측면 클로즈업으로 수리영의 감긴 눈과 어깨에 파묻힌 얼굴의 조용한 의존감을 강조한다.", "prompt_en": "Photograph the moment in an intimate eye-level side-profile close-up, with Suriyoung’s closed eyes and the contact of her nose against Minsuk’s shoulder as the precise focal point. Let Minsuk’s shoulder and apron anchor the frame while the apartment falls into soft, shallow-focus depth; the narrow morning sunlight grazes their profiles with restrained natural contrast." }, { "approach_ko": "거리를 둔 정적인 측면 환경 투숏으로 두 사람의 밀착과 생활감 있는 옥탑방 공간을 함께 담는다.", "prompt_en": "Use a restrained medium-wide side-on composition that holds both women within the compact apartment and preserves the full spatial context around their quiet contact. Keep their overlapping profiles visually concentrated against the deeper room, with the dining and kitchen areas remaining legible in layered depth; allow the narrow window light to cut gently through the morning interior while the rest of the room stays softly subdued." } ] }, "reused": false }, "S5sh9": { "input_fingerprint": "4515ac1c5374d63e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 감은 채 민숙의 어깨에 코를 묻은 측면 구도\n\nLOCATION (lock): A compact rooftop apartment interior with a small living-room window, a dining table, a sink and stove area, and bedrooms opening off the main space. A pot boils beside herbs being washed at the faucet. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 still wearing her apron, and the small circular mark remains on her 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 감은 채 민숙의 어깨에 코를 묻은 측면 구도\n\nLOCATION (lock): A compact rooftop apartment interior with a small living-room window, a dining table, a sink and stove area, and bedrooms opening off the main space. A pot boils beside herbs being washed at the faucet. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 still wearing her apron, and the small circular mark remains on her 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in an intimate eye-level side-profile close-up, with Suriyoung’s closed eyes and the contact of her nose against Minsuk’s shoulder as the precise focal point. Let Minsuk’s shoulder and apron anchor the frame while the apartment falls into soft, shallow-focus depth; the narrow morning sunlight grazes their profiles with restrained natural 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 감은 채 민숙의 어깨에 코를 묻은 측면 구도\n\nLOCATION (lock): A compact rooftop apartment interior with a small living-room window, a dining table, a sink and stove area, and bedrooms opening off the main space. A pot boils beside herbs being washed at the faucet. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 still wearing her apron, and the small circular mark remains on her 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained medium-wide side-on composition that holds both women within the compact apartment and preserves the full spatial context around their quiet contact. Keep their overlapping profiles visually concentrated against the deeper room, with the dining and kitchen areas remaining legible in layered depth; allow the narrow window light to cut gently through the morning interior while the rest of the room stays softly subdued.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 감은 채 민숙의 어깨에 코를 묻은 측면 구도\n\nLOCATION (lock): A compact rooftop apartment interior with a small living-room window, a dining table, a sink and stove area, and bedrooms opening off the main space. A pot boils beside herbs being washed at the faucet. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 still wearing her apron, and the small circular mark remains on her 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in an intimate eye-level side-profile close-up, with Suriyoung’s closed eyes and the contact of her nose against Minsuk’s shoulder as the precise focal point. Let Minsuk’s shoulder and apron anchor the frame while the apartment falls into soft, shallow-focus depth; the narrow morning sunlight grazes their profiles with restrained natural 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 수리영이 눈을 감은 채 민숙의 어깨에 코를 묻은 측면 구도\n\nLOCATION (lock): A compact rooftop apartment interior with a small living-room window, a dining table, a sink and stove area, and bedrooms opening off the main space. A pot boils beside herbs being washed at the faucet. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 still wearing her apron, and the small circular mark remains on her 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained medium-wide side-on composition that holds both women within the compact apartment and preserves the full spatial context around their quiet contact. Keep their overlapping profiles visually concentrated against the deeper room, with the dining and kitchen areas remaining legible in layered depth; allow the narrow window light to cut gently through the morning interior while the rest of the room stays softly subdued.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S5sh9.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 민숙: 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S5sh9.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 민숙: 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 민숙: 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A", "C", "D" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "측면 구도와 인물의 행동은 잘 구현되었으나, 배경의 싱크대를 창문 아래로 임의 이동시켜 공간 기준을 위반함." }, { "label": "B", "score": 8, "verdict_ko": "지시된 측면 구도와 인물의 자세를 정확히 묘사했으며, 기준 공간의 구조와 지정된 소품 위치를 완벽하게 유지함." }, { "label": "C", "score": 5, "verdict_ko": "공간 구조가 변형되었으며, 민숙의 손목에 있어야 할 표식이 수리영의 팔에 잘못 그려져 소품 상태 지시를 위반함." }, { "label": "D", "score": 4, "verdict_ko": "측면 묘사는 있으나 공간 기준 이미지의 카메라 앵글을 그대로 복사하여 금지 사항을 명백히 위반함." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A", "C", "D" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "측면 구도와 인물의 행동은 잘 구현되었으나, 배경의 싱크대를 창문 아래로 임의 이동시켜 공간 기준을 위반함." }, { "label": "B", "score": 8, "verdict_ko": "지시된 측면 구도와 인물의 자세를 정확히 묘사했으며, 기준 공간의 구조와 지정된 소품 위치를 완벽하게 유지함." }, { "label": "C", "score": 5, "verdict_ko": "공간 구조가 변형되었으며, 민숙의 손목에 있어야 할 표식이 수리영의 팔에 잘못 그려져 소품 상태 지시를 위반함." }, { "label": "D", "score": 4, "verdict_ko": "측면 묘사는 있으나 공간 기준 이미지의 카메라 앵글을 그대로 복사하여 금지 사항을 명백히 위반함." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "완벽한 측면 구도, 인물의 자세와 의상, 손목 표식 및 배경의 세부 동작(흐르는 물, 끓는 냄비)을 모두 정확히 구현했습니다." }, { "label": "C", "score": 5, "verdict_ko": "구도와 자세는 적절하나 수리영의 재킷이 누락되었고 배경의 세부 동작이 없습니다." }, { "label": "A", "score": 4, "verdict_ko": "구도는 무난하지만 표식의 위치가 손목이 아니며 배경의 디테일이 부족합니다." }, { "label": "B", "score": 3, "verdict_ko": "표식이 잘못된 인물(수리영)에게 적용되었으며 의상과 배경의 세부 묘사가 지침과 어긋납니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "완벽한 측면 구도, 인물의 자세와 의상, 손목 표식 및 배경의 세부 동작(흐르는 물, 끓는 냄비)을 모두 정확히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "구도와 자세는 적절하나 수리영의 재킷이 누락되었고 배경의 세부 동작이 없습니다." }, { "label": "D", "score": 4, "verdict_ko": "구도는 무난하지만 표식의 위치가 손목이 아니며 배경의 디테일이 부족합니다." }, { "label": "C", "score": 3, "verdict_ko": "표식이 잘못된 인물(수리영)에게 적용되었으며 의상과 배경의 세부 묘사가 지침과 어긋납니다." } ] }, "combined": { "totals": { "A": 13, "B": 13, "C": 8, "D": 8 }, "ranking": [ "A", "B", "C", "D" ], "agreement": false, "policy": 1 } }, "totals": { "A": 13, "B": 13, "C": 8, "D": 8 }, "selected": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "측면 구도와 인물의 행동은 잘 구현되었으나, 배경의 싱크대를 창문 아래로 임의 이동시켜 공간 기준을 위반함." }, { "label": "B", "score": 8, "verdict_ko": "지시된 측면 구도와 인물의 자세를 정확히 묘사했으며, 기준 공간의 구조와 지정된 소품 위치를 완벽하게 유지함." }, { "label": "C", "score": 5, "verdict_ko": "공간 구조가 변형되었으며, 민숙의 손목에 있어야 할 표식이 수리영의 팔에 잘못 그려져 소품 상태 지시를 위반함." }, { "label": "D", "score": 4, "verdict_ko": "측면 묘사는 있으나 공간 기준 이미지의 카메라 앵글을 그대로 복사하여 금지 사항을 명백히 위반함." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "거실 창문 아래에 위치 레퍼런스 사진에는 존재하지 않는 싱크대가 잘못 추가되었습니다.", "fix_en": "Remove the hallucinated sink and faucet from under the left window; move the herbs to the actual sink next to the stove in the background." }, { "issue_ko": "레이아웃 스케치와 반대로 두 인물의 위치와 시선 방향이 좌우로 완전히 뒤집혀 있습니다.", "fix_en": "Mirror the characters' placement and poses so that Minsuk stands on the left facing the window, with Suriyoung behind her on the right." }, { "issue_ko": "민숙이 레퍼런스에 있는 두툼한 베이지색 스웨터 대신 얇은 갈색 셔츠를 입고 있습니다.", "fix_en": "Change Minsuk's top to match the thick light beige crewneck sweater shown in her character reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the hallucinated sink and faucet from under the left window; move the herbs to the actual sink next to the stove in the background.\n- Mirror the characters' placement and poses so that Minsuk stands on the left facing the window, with Suriyoung behind her on the right.\n- Change Minsuk's top to match the thick light beige crewneck sweater shown in 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": true, "ref_mode": "플레이트+콘티+엔티티 (4택1: 콘티 승·A=변형0)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B02", "choice": "Candidate B", "confident": false, "reason_ko": "인물 간의 포옹 구도만 묘사되어 있으며 특정 공간(침실, 욕실 등)을 지시하는 단서가 없으므로 기존에 할당된 컷을 유지합니다.", "kept": "L05B02" } }, "S5sh10::variants": { "author_fp": "208331516c3ebe15", "author": { "variants": [ { "approach_ko": "손목을 비스듬히 파고드는 극근접 시점과 얕은 심도로 표식의 촉감과 긴장감을 강조한다.", "prompt_en": "Photograph the wrist in an intimate, oblique extreme close-up, with the small circular mark held in precise critical focus and the surrounding skin falling gently out of focus. Let the narrow morning sunlight rake across the wrist, revealing natural skin texture and giving the mark quiet dramatic emphasis; reduce the warm apartment interior and apron to soft, unobtrusive color fields at the frame edges." }, { "approach_ko": "손목 피부와 카메라를 거의 평행하게 맞춘 정면적이고 절제된 클로즈업으로 표식을 명료하게 기록한다.", "prompt_en": "Frame the wrist in a restrained, nearly perpendicular close-up, presenting the skin surface and circular mark with direct, clinical clarity rather than sensual softness. Use a steadier depth of field so the wrist contour and mark remain crisply legible together, with the narrow window light creating a simple directional division between light and shade and the composition kept spare and geometrically controlled." } ] }, "reused": false }, "S5sh10": { "input_fingerprint": "8e1143c06bd5d383", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 민숙의 손목 피부에 새겨진 작은 원 모양의 표식이 뚜렷하게 보이는 클로즈업\n\nLOCATION (lock): A compact rooftop apartment interior with a small living-room window, a dining table, a sink and stove area, and bedrooms opening off the main space. A pot boils beside herbs being washed at the faucet. 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 daylight interior look, warm domestic materials, and the woman's continuing clothing and arm position from the reference. Exclude the younger woman's face and the embrace from the close frame, showing only the wrist and its small circular mark.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 same small circular mark remains clearly defined on Minsuk’s wrist; she is still wearing her apron.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 민숙의 손목 피부에 새겨진 작은 원 모양의 표식이 뚜렷하게 보이는 클로즈업\n\nLOCATION (lock): A compact rooftop apartment interior with a small living-room window, a dining table, a sink and stove area, and bedrooms opening off the main space. A pot boils beside herbs being washed at the faucet. 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 daylight interior look, warm domestic materials, and the woman's continuing clothing and arm position from the reference. Exclude the younger woman's face and the embrace from the close frame, showing only the wrist and its small circular mark.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 same small circular mark remains clearly defined on Minsuk’s wrist; she is still wearing her apron.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the wrist in an intimate, oblique extreme close-up, with the small circular mark held in precise critical focus and the surrounding skin falling gently out of focus. Let the narrow morning sunlight rake across the wrist, revealing natural skin texture and giving the mark quiet dramatic emphasis; reduce the warm apartment interior and apron to soft, unobtrusive color fields at the frame edges.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, narrow sunlight through a small window.\n\nSHOT TEXT (authoritative, Korean): 민숙의 손목 피부에 새겨진 작은 원 모양의 표식이 뚜렷하게 보이는 클로즈업\n\nLOCATION (lock): A compact rooftop apartment interior with a small living-room window, a dining table, a sink and stove area, and bedrooms opening off the main space. A pot boils beside herbs being washed at the faucet. 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 daylight interior look, warm domestic materials, and the woman's continuing clothing and arm position from the reference. Exclude the younger woman's face and the embrace from the close frame, showing only the wrist and its small circular mark.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 same small circular mark remains clearly defined on Minsuk’s wrist; she is still wearing her apron.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame the wrist in a restrained, nearly perpendicular close-up, presenting the skin surface and circular mark with direct, clinical clarity rather than sensual softness. Use a steadier depth of field so the wrist contour and mark remain crisply legible together, with the narrow window light creating a simple directional division between light and shade and the composition kept spare and geometrically controlled.\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": "레퍼런스의 팔 자세(아래로 향함)와 의상을 정확히 유지하며 손목의 표식을 클로즈업으로 잘 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "팔을 수평으로 들어 올려 레퍼런스의 팔 자세를 유지하라는 지침을 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S5sh9_sel.png" }, { "label": "CHARACTER REFERENCE — 민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "이전 샷의 연속성을 어기고 팔을 앞으로 뻗어 손등이 위를 향하게 들고 있습니다.", "fix_en": "Point the arm downward to match the relaxed, hanging position established in the reference." }, { "issue_ko": "참조 이미지에서는 표식이 손목 안쪽 측면(엄지 쪽)에 있으나, 생성된 이미지에서는 손목 바깥쪽(등)에 있습니다.", "fix_en": "Move the circular mark to the side of the wrist below the thumb." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Point the arm downward to match the relaxed, hanging position established in the reference.\n- Move the circular mark to the side of the wrist below the thumb.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S5sh9" } }, "S6sh1::variants": { "author_fp": "e303ca52d08df144", "author": { "variants": [ { "approach_ko": "눈높이 병행 트래킹으로 인물의 선명한 전신 측면과 흐르는 해안 배경을 대비한다.", "prompt_en": "Use an eye-level lateral tracking view, holding her complete profile and bicycle crisply in a balanced medium-wide frame with generous lead room ahead. Let the coastal background carry restrained horizontal motion blur while her pedaling body, face, and wind-swept hair remain sharply legible, giving the instant fluid forward momentum. Natural daylight with clean, dimensional modeling and realistic contrast." }, { "approach_ko": "낮은 위치의 넓은 고정 구도로 도로의 진행선과 탁 트인 바다 속 전진감을 강조한다.", "prompt_en": "Photograph from a low, wider vantage near road level, keeping her entire figure and bicycle relatively compact within an expansive coastal composition. Use the road’s directional geometry and layered depth toward the open water to emphasize travel, with clear negative space in front of her and the mid-action pose frozen cleanly rather than blurred. Shape the daylight softly across her profile while preserving depth and detail throughout the setting." } ] }, "reused": false }, "S6sh1": { "input_fingerprint": "ba1389a50419db66", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 해안가 도로를 배경으로 자전거 페달을 밟으며 앞으로 나아가는 수리영의 전신 측면, 머리카락이 뒤로 휘날리는 mid-action\n\nLOCATION (lock): A coastal road running alongside the sea, wide enough for a bicycle route with open views toward the water. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong continues riding her bicycle with her earphones in place.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 해안가 도로를 배경으로 자전거 페달을 밟으며 앞으로 나아가는 수리영의 전신 측면, 머리카락이 뒤로 휘날리는 mid-action\n\nLOCATION (lock): A coastal road running alongside the sea, wide enough for a bicycle route with open views toward the water. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong continues riding her bicycle with her earphones in place.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an eye-level lateral tracking view, holding her complete profile and bicycle crisply in a balanced medium-wide frame with generous lead room ahead. Let the coastal background carry restrained horizontal motion blur while her pedaling body, face, and wind-swept hair remain sharply legible, giving the instant fluid forward momentum. Natural daylight with clean, dimensional modeling and realistic 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 해안가 도로를 배경으로 자전거 페달을 밟으며 앞으로 나아가는 수리영의 전신 측면, 머리카락이 뒤로 휘날리는 mid-action\n\nLOCATION (lock): A coastal road running alongside the sea, wide enough for a bicycle route with open views toward the water. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong continues riding her bicycle with her earphones in place.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a low, wider vantage near road level, keeping her entire figure and bicycle relatively compact within an expansive coastal composition. Use the road’s directional geometry and layered depth toward the open water to emphasize travel, with clear negative space in front of her and the mid-action pose frozen cleanly rather than blurred. Shape the daylight softly across her profile while preserving depth and detail throughout the setting.\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": 7 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "전신 측면 구도, 휘날리는 머리카락의 역동성, 이어폰 착용 상태 및 레퍼런스 인물과의 싱크로율이 모두 우수하게 구현되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "지정된 구도와 배경 조건은 잘 충족했으나, 피사체가 카메라에서 다소 멀어 인물의 디테일과 액션의 역동성이 A에 비해 떨어집니다." } ], "refs": [ { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "자전거 프레임(다운튜브)에 텍스트가 포함되어 있습니다.", "fix_en": "Remove all text from the bicycle frame." }, { "issue_ko": "자전거 크랭크 암의 각도가 비정상적(180도가 아님)이며, 페달 부근에 세 개의 발(신발)이 그려져 있습니다.", "fix_en": "Redraw the bicycle crankset so the two arms are exactly 180 degrees apart, and ensure only two legs and two feet are visible on the pedals." } ] }, "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 bicycle frame.\n- Redraw the bicycle crankset so the two arms are exactly 180 degrees apart, and ensure only two legs and two feet are visible on the pedals.\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", "share_plan": { "ref_plan": "background" } }, "S7sh2::variants": { "author_fp": "f48d1defba93ff47", "author": { "variants": [ { "approach_ko": "철망과 개의 흐릿한 전경 너머 수리영의 정면 응시를 대칭적으로 압축한 밀착 클로즈업.", "prompt_en": "Photograph this as a tight, eye-level frontal close-up with rigorous symmetry, compressing the wire mesh against her face while keeping it visually distinct from her skin. Let the Rottweiler register only as a soft, dark foreground presence at the lower edge, directing all attention to her sharply focused eyes and controlled facial tension. Use shallow depth and restrained natural daytime light for an intimate, confrontational stillness." }, { "approach_ko": "로트와일러의 눈높이에서 개와 철망을 강한 전경으로 두고 수리영의 정면 얼굴을 깊이감 있게 포착한 구도.", "prompt_en": "Shoot from a lower, Rottweiler-eye-height position in a wider layered composition, with the dog’s head and eyes forming a forceful near foreground, the wire mesh defining the middle plane, and her frontal face held beyond it. Use deeper focus so the locked exchange of gazes and the physical separation remain simultaneously legible, while the enclosure structure creates firm converging lines around her. Shape the available daytime light with stronger contrast across the planes, giving the moment a tense, watchful spatial realism." } ] }, "reused": false }, "S7sh2": { "input_fingerprint": "1aa91e948e7b79b5", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철망 너머로 로트와일러의 눈을 매섭게 노려보는 수리영의 정면 얼굴\n\nLOCATION (lock): A dog shelter with secure enclosures for multiple dogs and a work area for vaccinations, grooming, ointment application, and bathing. 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철망 너머로 로트와일러의 눈을 매섭게 노려보는 수리영의 정면 얼굴\n\nLOCATION (lock): A dog shelter with secure enclosures for multiple dogs and a work area for vaccinations, grooming, ointment application, and bathing. 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph this as a tight, eye-level frontal close-up with rigorous symmetry, compressing the wire mesh against her face while keeping it visually distinct from her skin. Let the Rottweiler register only as a soft, dark foreground presence at the lower edge, directing all attention to her sharply focused eyes and controlled facial tension. Use shallow depth and restrained natural daytime light for an intimate, confrontational 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철망 너머로 로트와일러의 눈을 매섭게 노려보는 수리영의 정면 얼굴\n\nLOCATION (lock): A dog shelter with secure enclosures for multiple dogs and a work area for vaccinations, grooming, ointment application, and bathing. 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from a lower, Rottweiler-eye-height position in a wider layered composition, with the dog’s head and eyes forming a forceful near foreground, the wire mesh defining the middle plane, and her frontal face held beyond it. Use deeper focus so the locked exchange of gazes and the physical separation remain simultaneously legible, while the enclosure structure creates firm converging lines around her. Shape the available daytime light with stronger contrast across the planes, giving the moment a tense, watchful spatial 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.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철망 너머로 로트와일러의 눈을 매섭게 노려보는 수리영의 정면 얼굴\n\nLOCATION (lock): A dog shelter with secure enclosures for multiple dogs and a work area for vaccinations, grooming, ointment application, and bathing. 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph this as a tight, eye-level frontal close-up with rigorous symmetry, compressing the wire mesh against her face while keeping it visually distinct from her skin. Let the Rottweiler register only as a soft, dark foreground presence at the lower edge, directing all attention to her sharply focused eyes and controlled facial tension. Use shallow depth and restrained natural daytime light for an intimate, confrontational 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철망 너머로 로트와일러의 눈을 매섭게 노려보는 수리영의 정면 얼굴\n\nLOCATION (lock): A dog shelter with secure enclosures for multiple dogs and a work area for vaccinations, grooming, ointment application, and bathing. 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from a lower, Rottweiler-eye-height position in a wider layered composition, with the dog’s head and eyes forming a forceful near foreground, the wire mesh defining the middle plane, and her frontal face held beyond it. Use deeper focus so the locked exchange of gazes and the physical separation remain simultaneously legible, while the enclosure structure creates firm converging lines around her. Shape the available daytime light with stronger contrast across the planes, giving the moment a tense, watchful spatial 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." }, "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/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": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'수리영의 정면 얼굴'과 '철망 너머로 로트와일러의 눈을 노려보는' 샷 텍스트를 지정된 로케이션과 함께 가장 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "얼굴과 액션은 무난하게 표현되었으나, 배경의 우리 구조물이 원본 로케이션과 반대인 우측 골목에 배치되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "지정된 야외 로케이션 구조를 무시한 실내 공간이며, 개의 뒷모습만 보여 텍스트의 시선 교환 액션을 명확히 확인할 수 없습니다." }, { "label": "D", "score": 4, "verdict_ko": "요구된 철망 대신 쇠창살이 생성되었고, 개가 카메라를 향하고 있어 개의 눈을 노려보는 핵심 액션이 성립하지 않습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'수리영의 정면 얼굴'과 '철망 너머로 로트와일러의 눈을 노려보는' 샷 텍스트를 지정된 로케이션과 함께 가장 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "얼굴과 액션은 무난하게 표현되었으나, 배경의 우리 구조물이 원본 로케이션과 반대인 우측 골목에 배치되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "지정된 야외 로케이션 구조를 무시한 실내 공간이며, 개의 뒷모습만 보여 텍스트의 시선 교환 액션을 명확히 확인할 수 없습니다." }, { "label": "D", "score": 4, "verdict_ko": "요구된 철망 대신 쇠창살이 생성되었고, 개가 카메라를 향하고 있어 개의 눈을 노려보는 핵심 액션이 성립하지 않습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "철망 너머 인물의 정면 얼굴과 로트와일러와의 시선 교환, 지정된 장소의 배경까지 샷 텍스트를 가장 완벽하게 구현함." }, { "label": "C", "score": 6, "verdict_ko": "장소 레퍼런스는 잘 반영했으나 인물의 얼굴이 완전한 정면이 아니며 개가 다른 곳을 보고 있어 시선 교환이 이루어지지 않음." }, { "label": "A", "score": 5, "verdict_ko": "지정된 장소는 일치하지만 샷이 너무 넓어 '정면 얼굴'의 의도가 흐려졌고 개가 카메라를 응시함." }, { "label": "B", "score": 3, "verdict_ko": "제공된 유일한 야외 장소 레퍼런스를 완전히 무시하고 실내 공간을 임의로 생성한 심각한 배경 위반임." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "철망 너머 인물의 정면 얼굴과 로트와일러와의 시선 교환, 지정된 장소의 배경까지 샷 텍스트를 가장 완벽하게 구현함." }, { "label": "B", "score": 6, "verdict_ko": "장소 레퍼런스는 잘 반영했으나 인물의 얼굴이 완전한 정면이 아니며 개가 다른 곳을 보고 있어 시선 교환이 이루어지지 않음." }, { "label": "D", "score": 5, "verdict_ko": "지정된 장소는 일치하지만 샷이 너무 넓어 '정면 얼굴'의 의도가 흐려졌고 개가 카메라를 응시함." }, { "label": "C", "score": 3, "verdict_ko": "제공된 유일한 야외 장소 레퍼런스를 완전히 무시하고 실내 공간을 임의로 생성한 심각한 배경 위반임." } ] }, "combined": { "totals": { "A": 14, "B": 12, "C": 6, "D": 9 }, "ranking": [ "A", "B", "D", "C" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 12, "C": 6, "D": 9 }, "selected": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'수리영의 정면 얼굴'과 '철망 너머로 로트와일러의 눈을 노려보는' 샷 텍스트를 지정된 로케이션과 함께 가장 정확하게 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "얼굴과 액션은 무난하게 표현되었으나, 배경의 우리 구조물이 원본 로케이션과 반대인 우측 골목에 배치되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "지정된 야외 로케이션 구조를 무시한 실내 공간이며, 개의 뒷모습만 보여 텍스트의 시선 교환 액션을 명확히 확인할 수 없습니다." }, { "label": "D", "score": 4, "verdict_ko": "요구된 철망 대신 쇠창살이 생성되었고, 개가 카메라를 향하고 있어 개의 눈을 노려보는 핵심 액션이 성립하지 않습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/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": "레이아웃 스케치에 따라 개가 화면 왼쪽 아래에 배치되어야 하지만, 오른쪽 아래에 있습니다.", "fix_en": "Move the Rottweiler to the bottom left foreground to match the layout sketch's placement." }, { "issue_ko": "샷 텍스트에서 로트와일러의 눈을 노려보라고 명시했으나, 수리영이 개가 아닌 카메라 정면을 응시하고 있습니다.", "fix_en": "Redirect Suri-young's eyes downward and to the side so she is glaring directly at the Rottweiler's eyes." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the Rottweiler to the bottom left foreground to match the layout sketch's placement.\n- Redirect Suri-young's eyes downward and to the side so she is glaring directly at the Rottweiler's eyes.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "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": "seed-bg+콘티+엔티티 (복잡구조물 4택1: 콘티 승·A=변형0)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready" }, "S8sh4::variants": { "author_fp": "dbe136175ccab7af", "author": { "variants": [ { "approach_ko": "수리영의 놀란 눈과 굳은 표정에 집중하는 타이트한 측면 클로즈업.", "prompt_en": "Frame a tight side-profile close-up at eye level, with Su-ri-yeong’s face dominating the image and her fixed eyeline aimed just beyond the frame toward the employee. Keep her anatomically natural eyes sharply focused and opened wide through performance, with restrained tension in her brow, lips, and jaw. Use shallow depth of field and soft dusk-neon spill to isolate the instant of startled recognition while the supermarket environment falls into subdued blur." }, { "approach_ko": "직원을 향한 시선 방향에 여백을 두고 마트의 깊이 속에 수리영을 배치하는 환경적 측면 숏.", "prompt_en": "Photograph Su-ri-yeong in a wider side-on composition from slightly below shoulder height, placing her to one side of the frame and leaving charged negative space in the direction of the employee she faces off camera. Preserve visible depth through the supermarket entrance, aisles, and stocking areas, with dusk and neon illumination layering the foreground and background. Keep her profile and widened, natural-eyed reaction clearly readable while the surrounding commercial bustle gives the confrontation a grounded, immediate scale." } ] }, "reused": false }, "S8sh4": { "input_fingerprint": "8cec286d7788f852", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 남자 직원과 마주 본 상태로 눈을 동그랗게 뜬 수리영의 측면\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains parked outside the mart, and she still has her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 남자 직원과 마주 본 상태로 눈을 동그랗게 뜬 수리영의 측면\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains parked outside the mart, and she still has her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight side-profile close-up at eye level, with Su-ri-yeong’s face dominating the image and her fixed eyeline aimed just beyond the frame toward the employee. Keep her anatomically natural eyes sharply focused and opened wide through performance, with restrained tension in her brow, lips, and jaw. Use shallow depth of field and soft dusk-neon spill to isolate the instant of startled recognition while the supermarket environment falls into subdued 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 남자 직원과 마주 본 상태로 눈을 동그랗게 뜬 수리영의 측면\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains parked outside the mart, and she still has her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph Su-ri-yeong in a wider side-on composition from slightly below shoulder height, placing her to one side of the frame and leaving charged negative space in the direction of the employee she faces off camera. Preserve visible depth through the supermarket entrance, aisles, and stocking areas, with dusk and neon illumination layering the foreground and background. Keep her profile and widened, natural-eyed reaction clearly readable while the surrounding commercial bustle gives the confrontation a grounded, immediate scale.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 남자 직원과 마주 본 상태로 눈을 동그랗게 뜬 수리영의 측면\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains parked outside the mart, and she still has her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight side-profile close-up at eye level, with Su-ri-yeong’s face dominating the image and her fixed eyeline aimed just beyond the frame toward the employee. Keep her anatomically natural eyes sharply focused and opened wide through performance, with restrained tension in her brow, lips, and jaw. Use shallow depth of field and soft dusk-neon spill to isolate the instant of startled recognition while the supermarket environment falls into subdued 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.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 남자 직원과 마주 본 상태로 눈을 동그랗게 뜬 수리영의 측면\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains parked outside the mart, and she still has her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph Su-ri-yeong in a wider side-on composition from slightly below shoulder height, placing her to one side of the frame and leaving charged negative space in the direction of the employee she faces off camera. Preserve visible depth through the supermarket entrance, aisles, and stocking areas, with dusk and neon illumination layering the foreground and background. Keep her profile and widened, natural-eyed reaction clearly readable while the surrounding commercial bustle gives the confrontation a grounded, immediate scale.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S8sh4.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S8sh4.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B01.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", "B", "C", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "지시된 측면 구도와 직원과의 대면 상태를 정확히 연출했으며, 외부에 주차된 자전거 등 디테일한 설정도 모두 충족했습니다." }, { "label": "B", "score": 6, "verdict_ko": "측면 구도와 상호작용은 훌륭히 포착했으나, 마트 밖에 있어야 할 자전거가 내부에 배치된 위치 오류가 있습니다." }, { "label": "C", "score": 5, "verdict_ko": "직원과 마주보는 상황은 표현되었으나, 텍스트가 요구한 '측면' 앵글이 아닌 정면에 가까운 반측면으로 촬영되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "수리영의 측면에 가까운 구도지만, 마주 보아야 할 남자 직원이 전면에 존재하지 않아 핵심적인 상호작용이 누락되었습니다." } ] }, "forward_normalized": { "winner": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "지시된 측면 구도와 직원과의 대면 상태를 정확히 연출했으며, 외부에 주차된 자전거 등 디테일한 설정도 모두 충족했습니다." }, { "label": "B", "score": 6, "verdict_ko": "측면 구도와 상호작용은 훌륭히 포착했으나, 마트 밖에 있어야 할 자전거가 내부에 배치된 위치 오류가 있습니다." }, { "label": "C", "score": 5, "verdict_ko": "직원과 마주보는 상황은 표현되었으나, 텍스트가 요구한 '측면' 앵글이 아닌 정면에 가까운 반측면으로 촬영되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "수리영의 측면에 가까운 구도지만, 마주 보아야 할 남자 직원이 전면에 존재하지 않아 핵심적인 상호작용이 누락되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A", "C", "D" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "치명적인 오류는 없으나 샷이 너무 넓어 눈을 동그랗게 뜬 측면이라는 핵심 프레이밍 지시를 약화시켰습니다." }, { "label": "B", "score": 7, "verdict_ko": "지시된 측면 구도와 놀란 표정을 가장 잘 살렸으나, 마주 보는 남자의 복장(마트 직원) 묘사가 다소 부족합니다." }, { "label": "C", "score": 3, "verdict_ko": "직원과의 대치 상황은 훌륭하나 화면 하단에 1인칭 시점의 제3자 손이 등장하는 치명적 오류와 자전거가 실내에 있는 오류가 있습니다." }, { "label": "D", "score": 2, "verdict_ko": "직원과 마주보는 상황을 연출하지 못했으며, 화면 하단에 제3자의 손이 등장하는 치명적인 구조 오류가 있습니다." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "D", "B", "A" ], "verdicts": [ { "label": "D", "score": 5, "verdict_ko": "치명적인 오류는 없으나 샷이 너무 넓어 눈을 동그랗게 뜬 측면이라는 핵심 프레이밍 지시를 약화시켰습니다." }, { "label": "C", "score": 7, "verdict_ko": "지시된 측면 구도와 놀란 표정을 가장 잘 살렸으나, 마주 보는 남자의 복장(마트 직원) 묘사가 다소 부족합니다." }, { "label": "B", "score": 3, "verdict_ko": "직원과의 대치 상황은 훌륭하나 화면 하단에 1인칭 시점의 제3자 손이 등장하는 치명적 오류와 자전거가 실내에 있는 오류가 있습니다." }, { "label": "A", "score": 2, "verdict_ko": "직원과 마주보는 상황을 연출하지 못했으며, 화면 하단에 제3자의 손이 등장하는 치명적인 구조 오류가 있습니다." } ] }, "combined": { "totals": { "A": 6, "B": 9, "C": 12, "D": 12 }, "ranking": [ "C", "D", "B", "A" ], "agreement": false, "policy": 1 } }, "totals": { "A": 6, "B": 9, "C": 12, "D": 12 }, "selected": "C", "ranking": [ "C", "D", "B", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "지시된 측면 구도와 직원과의 대면 상태를 정확히 연출했으며, 외부에 주차된 자전거 등 디테일한 설정도 모두 충족했습니다." }, { "label": "B", "score": 6, "verdict_ko": "측면 구도와 상호작용은 훌륭히 포착했으나, 마트 밖에 있어야 할 자전거가 내부에 배치된 위치 오류가 있습니다." }, { "label": "C", "score": 5, "verdict_ko": "직원과 마주보는 상황은 표현되었으나, 텍스트가 요구한 '측면' 앵글이 아닌 정면에 가까운 반측면으로 촬영되었습니다." }, { "label": "A", "score": 4, "verdict_ko": "수리영의 측면에 가까운 구도지만, 마주 보아야 할 남자 직원이 전면에 존재하지 않아 핵심적인 상호작용이 누락되었습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프롬프트의 인물 제한 규칙(수리영 외 인물 등장 불가)과 시각적 접근(화면 밖을 응시)을 어기고 화면 왼쪽에 남성 직원이 크게 나타나 있습니다.", "fix_en": "Remove the man from the left foreground completely so that only Su-ri-yeong remains in the frame, keeping her eyeline directed just beyond the edge of the screen." }, { "issue_ko": "텍스트 금지 규칙을 위반하여 배경의 핑크색 네온사인과 바닥의 종이 상자들에 문자가 포함되어 있습니다.", "fix_en": "Remove all letters and text from the pink neon sign and the cardboard boxes in the background, leaving them blank." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the man from the left foreground completely so that only Su-ri-yeong remains in the frame, keeping her eyeline directed just beyond the edge of the screen.\n- Remove all letters and text from the pink neon sign and the cardboard boxes in the background, leaving them blank.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L09B01", "B": "L09B02", "C": "L09B03" }, "assigned": "L09B01", "choice": "Candidate A", "confident": true, "reason_ko": "마트 직원과 마주 보고 있다는 설명으로 보아 마트 내부 공간이 배경이어야 하므로, 마트 안쪽을 보여주는 Candidate A가 가장 적합합니다.", "kept": "L09B01" } }, "S8sh5::variants": { "author_fp": "ab307f036ee47613", "author": { "variants": [ { "approach_ko": "정면 눈높이의 밀도 높은 상반신 클로즈업으로 통화에 집중한 수리영을 네온 배경에서 고립시킨다.", "prompt_en": "Frame a tight eye-level medium close-up from outside, with her upper torso dominant and the automatic doorway forming a restrained, nearly symmetrical boundary behind her. Use shallow depth of field so the dusk street and neon resolve into soft colored highlights, while mixed storefront spill and cool ambient light model her face naturally. Keep the visual emphasis on her concentrated listening posture and the phone held correctly against her ear." }, { "approach_ko": "마트 안쪽에서 자동문 너머를 비스듬히 바라보며 유리와 내외부 공간의 층으로 수리영의 고립감을 드러낸다.", "prompt_en": "Photograph her obliquely from just inside the supermarket, cropping to the upper body and holding her in a clean three-quarter profile beyond the automatic door. Use the glass, door frame, and receding interior and exterior planes to create layered depth, with faint neon reflections adding visual tension without obscuring her face. Favor a moderately deep focus and balanced dusk-to-interior illumination so the threshold itself becomes the compositional emphasis." } ] }, "reused": false }, "S8sh5": { "input_fingerprint": "8032bc43c6a1433e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 자동문 밖에 서서 휴대폰을 귀에 댄 수리영의 상체\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Her bicycle remains outside the mart while Su-ri-yeong holds her mobile phone to her ear and again tries to reach Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 자동문 밖에 서서 휴대폰을 귀에 댄 수리영의 상체\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Her bicycle remains outside the mart while Su-ri-yeong holds her mobile phone to her ear and again tries to reach Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight eye-level medium close-up from outside, with her upper torso dominant and the automatic doorway forming a restrained, nearly symmetrical boundary behind her. Use shallow depth of field so the dusk street and neon resolve into soft colored highlights, while mixed storefront spill and cool ambient light model her face naturally. Keep the visual emphasis on her concentrated listening posture and the phone held correctly against her ear.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 자동문 밖에 서서 휴대폰을 귀에 댄 수리영의 상체\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Her bicycle remains outside the mart while Su-ri-yeong holds her mobile phone to her ear and again tries to reach Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her obliquely from just inside the supermarket, cropping to the upper body and holding her in a clean three-quarter profile beyond the automatic door. Use the glass, door frame, and receding interior and exterior planes to create layered depth, with faint neon reflections adding visual tension without obscuring her face. Favor a moderately deep focus and balanced dusk-to-interior illumination so the threshold itself becomes the compositional emphasis.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 자동문 밖에 서서 휴대폰을 귀에 댄 수리영의 상체\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Her bicycle remains outside the mart while Su-ri-yeong holds her mobile phone to her ear and again tries to reach Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight eye-level medium close-up from outside, with her upper torso dominant and the automatic doorway forming a restrained, nearly symmetrical boundary behind her. Use shallow depth of field so the dusk street and neon resolve into soft colored highlights, while mixed storefront spill and cool ambient light model her face naturally. Keep the visual emphasis on her concentrated listening posture and the phone held correctly against her ear.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, neon signs visible.\n\nSHOT TEXT (authoritative, Korean): 마트 자동문 밖에 서서 휴대폰을 귀에 댄 수리영의 상체\n\nLOCATION (lock): A busy commercial district beside a harbor, crowded with neon signs, taxis, street vendors, and food tents, leading into a large supermarket with merchandise aisles and stocking areas. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Her bicycle remains outside the mart while Su-ri-yeong holds her mobile phone to her ear and again tries to reach Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her obliquely from just inside the supermarket, cropping to the upper body and holding her in a clean three-quarter profile beyond the automatic door. Use the glass, door frame, and receding interior and exterior planes to create layered depth, with faint neon reflections adding visual tension without obscuring her face. Favor a moderately deep focus and balanced dusk-to-interior illumination so the threshold itself becomes the compositional emphasis.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S8sh5.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S8sh5.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B02.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": 8, "verdict_ko": "'상체' 프레이밍을 정확히 구현했고 카메라 방향에 따른 장소 구조와 인물이 레퍼런스와 완벽히 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "샷의 크기가 '상체'가 아닌 무릎 위까지 넓어져 프레이밍 지시를 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "마트 내부가 있어야 할 유리문 너머로 거리가 정방향으로 나타나 물리적인 공간 왜곡(Hard Violation)이 발생했습니다." }, { "label": "D", "score": 5, "verdict_ko": "상체 프레이밍은 준수하나 마트 건물과 거리의 건축적 구조 배치가 레퍼런스와 심하게 다릅니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "D", "B", "C" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "'상체' 프레이밍을 정확히 구현했고 카메라 방향에 따른 장소 구조와 인물이 레퍼런스와 완벽히 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "샷의 크기가 '상체'가 아닌 무릎 위까지 넓어져 프레이밍 지시를 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "마트 내부가 있어야 할 유리문 너머로 거리가 정방향으로 나타나 물리적인 공간 왜곡(Hard Violation)이 발생했습니다." }, { "label": "D", "score": 5, "verdict_ko": "상체 프레이밍은 준수하나 마트 건물과 거리의 건축적 구조 배치가 레퍼런스와 심하게 다릅니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "D", "score": 8, "verdict_ko": "'자동문 밖 상체' 프레이밍을 정확히 구현했고, 장소의 입체적 구조와 지정된 의상을 완벽하게 일치시킴." }, { "label": "B", "score": 6, "verdict_ko": "지정된 프레이밍과 포즈는 충족했으나, 캐릭터의 재킷 색상과 디자인이 레퍼런스와 명백히 다름." }, { "label": "A", "score": 5, "verdict_ko": "의상과 포즈는 우수하지만, 자동문 '밖'이 아닌 마트 내부에 서 있어 공간 지시를 위반함." }, { "label": "C", "score": 4, "verdict_ko": "상체 샷 기준보다 앵글이 넓고 문 안쪽에 서 있어 샷 텍스트의 핵심 지시에서 벗어남." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "'자동문 밖 상체' 프레이밍을 정확히 구현했고, 장소의 입체적 구조와 지정된 의상을 완벽하게 일치시킴." }, { "label": "C", "score": 6, "verdict_ko": "지정된 프레이밍과 포즈는 충족했으나, 캐릭터의 재킷 색상과 디자인이 레퍼런스와 명백히 다름." }, { "label": "D", "score": 5, "verdict_ko": "의상과 포즈는 우수하지만, 자동문 '밖'이 아닌 마트 내부에 서 있어 공간 지시를 위반함." }, { "label": "B", "score": 4, "verdict_ko": "상체 샷 기준보다 앵글이 넓고 문 안쪽에 서 있어 샷 텍스트의 핵심 지시에서 벗어남." } ] }, "combined": { "totals": { "A": 16, "B": 8, "C": 9, "D": 10 }, "ranking": [ "A", "D", "C", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 16, "B": 8, "C": 9, "D": 10 }, "selected": "A", "ranking": [ "A", "D", "C", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "'상체' 프레이밍을 정확히 구현했고 카메라 방향에 따른 장소 구조와 인물이 레퍼런스와 완벽히 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "샷의 크기가 '상체'가 아닌 무릎 위까지 넓어져 프레이밍 지시를 위반했습니다." }, { "label": "C", "score": 3, "verdict_ko": "마트 내부가 있어야 할 유리문 너머로 거리가 정방향으로 나타나 물리적인 공간 왜곡(Hard Violation)이 발생했습니다." }, { "label": "D", "score": 5, "verdict_ko": "상체 프레이밍은 준수하나 마트 건물과 거리의 건축적 구조 배치가 레퍼런스와 심하게 다릅니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "자동문이 인물 뒤에서 대칭적인 경계를 이루어야 한다는 시각적 접근(VISUAL APPROACH) 지시가 무시되고, 문이 화면 왼쪽에 치우쳐 있습니다.", "fix_en": "Reframe the background so the automatic doorway forms a nearly symmetrical boundary directly behind the character's torso." }, { "issue_ko": "샷 텍스트에 명시되지 않은 사람의 형상이 화면 왼쪽 마트 유리창 너머로 보입니다.", "fix_en": "Remove the unprompted human figure visible inside the mart through the glass on the left." }, { "issue_ko": "배경의 간판들에 글자가 포함되어 있어 'No text anywhere' 규칙을 위반했습니다.", "fix_en": "Remove all recognizable text characters from the neon signs in the background, leaving only abstract glowing shapes." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Reframe the background so the automatic doorway forms a nearly symmetrical boundary directly behind the character's torso.\n- Remove the unprompted human figure visible inside the mart through the glass on the left.\n- Remove all recognizable text characters from the neon signs in the background, leaving only abstract glowing shapes.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L09B01", "B": "L09B02", "C": "L09B03" }, "assigned": "L09B02", "choice": "B", "confident": true, "reason_ko": "샷 설명에 '마트 자동문 밖에 서서'라고 명시되어 있으므로, 마트 자동문 외부를 보여주는 서브공간이 필요합니다.", "chosen": "L09B02" } }, "S9sh1::variants": { "author_fp": "898a099f2e1f350e", "author": { "variants": [ { "approach_ko": "해안도로와 검은 바다를 넓게 두고 수리영의 전신 윤곽과 강한 페달 동작을 측면에서 포착한다.", "prompt_en": "Photograph the moment in a wide lateral profile from a low, level camera position, keeping her full figure and bicycle cleanly readable against the dark coastal road and sea. Use restrained depth compression and strong negative space ahead of her, with subtle night backlight defining her human profile, bent posture, driving leg, and wind-swept clothing while retaining natural live-action detail within the shadow." }, { "approach_ko": "자전거 가까이의 낮은 후방 사선 시점으로 페달의 힘과 앞으로 쏠린 자세, 바람의 속도감을 강조한다.", "prompt_en": "Place the camera very low and close in a rear three-quarter tracking view, using an expansive lens feel so the bicycle and forceful pedal stroke dominate the foreground while her lowered torso drives diagonally through the frame. Let the road and dark shoreline recede deeply behind her; shape the available night light into a narrow edge around her body and flying garment, preserving realistic texture rather than reducing her to a featureless black shape." } ] }, "reused": false }, "S9sh1": { "input_fingerprint": "ccf9e053aa1fbf6d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 해안도로를 배경으로 허리를 숙인 채 한쪽 무릎을 굽혀 페달을 강하게 밟아 누르고, 옷자락이 바람에 뒤로 날리는 수리영의 실루엣\n\nLOCATION (lock): A coastal road beside the dark sea, serving as a bicycle route back toward the residential district. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong continues home on her bicycle with her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 해안도로를 배경으로 허리를 숙인 채 한쪽 무릎을 굽혀 페달을 강하게 밟아 누르고, 옷자락이 바람에 뒤로 날리는 수리영의 실루엣\n\nLOCATION (lock): A coastal road beside the dark sea, serving as a bicycle route back toward the residential district. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong continues home on her bicycle with her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in a wide lateral profile from a low, level camera position, keeping her full figure and bicycle cleanly readable against the dark coastal road and sea. Use restrained depth compression and strong negative space ahead of her, with subtle night backlight defining her human profile, bent posture, driving leg, and wind-swept clothing while retaining natural live-action detail within the 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 해안도로를 배경으로 허리를 숙인 채 한쪽 무릎을 굽혀 페달을 강하게 밟아 누르고, 옷자락이 바람에 뒤로 날리는 수리영의 실루엣\n\nLOCATION (lock): A coastal road beside the dark sea, serving as a bicycle route back toward the residential district. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong continues home on her bicycle with her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera very low and close in a rear three-quarter tracking view, using an expansive lens feel so the bicycle and forceful pedal stroke dominate the foreground while her lowered torso drives diagonally through the frame. Let the road and dark shoreline recede deeply behind her; shape the available night light into a narrow edge around her body and flying garment, preserving realistic texture rather than reducing her to a featureless black 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": 6, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지정된 해안도로의 난간과 배경을 정확히 구현하고 페달을 밟는 동작을 무난히 표현했으나, 지문이 요구한 실루엣 연출은 누락되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "왼쪽 다리가 앞바퀴 쪽으로 뻗어 있는 등 자전거의 구조와 인체 자세가 물리적으로 불가능하며(치명적 오류), 기준 이미지의 난간도 보이지 않습니다." } ], "refs": [ { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영이 계속 소지하고 있어야 할 휴대폰이 화면에 보이지 않습니다.", "fix_en": "Add a mobile phone to her hand or make it clearly visible in her pocket." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add a mobile phone to her hand or make it clearly visible in her pocket.\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", "share_plan": { "ref_plan": "background" } }, "S10sh6::variants": { "author_fp": "a0acae29ebcef83e", "author": { "variants": [ { "approach_ko": "창문 반사를 화면의 중심으로 삼아 공격 순간을 압축적으로 포착하는 불안정한 근접 구도.", "prompt_en": "Photograph the window reflection as the dominant image in a tight, tense composition, with the physical figures compressed into layered reflected depth yet retaining readable human anatomy, clothing, and facial performance. Use a slightly canted eye-level viewpoint and a close, urgent lens feel; let the force of the movement cut diagonally through the frame while hard, grazing night illumination gives the reflected action sharp separation without turning either person into an abstract silhouette." }, { "approach_ko": "낮고 넓은 시점에서 실제 몸싸움과 창문 속 그림자를 함께 배치해 옥상의 공간감과 습격의 충격을 강조한다.", "prompt_en": "Use a low, wider viewpoint that holds the immediate physical struggle in the foreground while the window reflection occupies a strong secondary plane, echoing the attack through shadow and movement. Build pronounced foreground-to-background depth from the rooftop architecture, with the open doorway and fixed rooftop elements anchoring the composition; keep the illumination sparse and directional so the figures remain fully human and legible while the reflected attacker reads as ominous and unidentified." } ] }, "reused": false }, "S10sh6": { "input_fingerprint": "05ba2a7721f14628", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 창문에 비친 그림자 속, 정체불명의 습격자의 실루엣이 민숙을 덮쳐 어깨를 강하게 물어뜯는 찰나의 역동적인 모습\n\nLOCATION (lock): An exposed rooftop above an old residential building, fitted with a water tank, clotheslines, a drying platform covered with herbs, and a metal entrance door to a small rooftop apartment. 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 door remains slightly open after the intruder’s entry. Minsuk still bears the small circular wrist mark as the attacker bites her shoulder.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 창문에 비친 그림자 속, 정체불명의 습격자의 실루엣이 민숙을 덮쳐 어깨를 강하게 물어뜯는 찰나의 역동적인 모습\n\nLOCATION (lock): An exposed rooftop above an old residential building, fitted with a water tank, clotheslines, a drying platform covered with herbs, and a metal entrance door to a small rooftop apartment. 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 door remains slightly open after the intruder’s entry. Minsuk still bears the small circular wrist mark as the attacker bites her shoulder.\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 window reflection as the dominant image in a tight, tense composition, with the physical figures compressed into layered reflected depth yet retaining readable human anatomy, clothing, and facial performance. Use a slightly canted eye-level viewpoint and a close, urgent lens feel; let the force of the movement cut diagonally through the frame while hard, grazing night illumination gives the reflected action sharp separation without turning either person into an abstract 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 창문에 비친 그림자 속, 정체불명의 습격자의 실루엣이 민숙을 덮쳐 어깨를 강하게 물어뜯는 찰나의 역동적인 모습\n\nLOCATION (lock): An exposed rooftop above an old residential building, fitted with a water tank, clotheslines, a drying platform covered with herbs, and a metal entrance door to a small rooftop apartment. 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 door remains slightly open after the intruder’s entry. Minsuk still bears the small circular wrist mark as the attacker bites her shoulder.\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, wider viewpoint that holds the immediate physical struggle in the foreground while the window reflection occupies a strong secondary plane, echoing the attack through shadow and movement. Build pronounced foreground-to-background depth from the rooftop architecture, with the open doorway and fixed rooftop elements anchoring the composition; keep the illumination sparse and directional so the figures remain fully human and legible while the reflected attacker reads as ominous and unidentified.\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": 6, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "비유적인 '실루엣' 묘사를 실제 인물로 사실적으로 연출했으나, 지정된 건물의 창문 크기와 형태를 임의로 변경한 점이 아쉽습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스의 카메라 구도를 금지 규정에 반해 그대로 복사했으며, 비유적 표현을 벽에 떠 있는 비현실적인 그림자로 직역하여 심각한 위반을 범했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L04B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/background_chain/seed_bg_rooftop_villa_complex_sel.png" } ], "critique": { "issues": [ { "issue_ko": "인물들이 창문에 비친 반사된 모습으로 연출되지 않고, 창문 너머 방 안에 있는 상태로 묘사되었습니다.", "fix_en": "Render the figures entirely as a reflection on the exterior of the window glass, rather than showing them inside the room." }, { "issue_ko": "옥탑방 문이 '살짝 열린' 상태를 유지해야 한다는 지시와 달리 너무 활짝 열려 있습니다.", "fix_en": "Adjust the rooftop room door so it is only slightly open." }, { "issue_ko": "문의 재질, 색상 및 디자인이 레퍼런스 사진의 갈색 금속 문과 일치하지 않고 바깥쪽으로 열려 있습니다.", "fix_en": "Modify the door's design, material, and color to exactly match the brown metal door in the location reference photograph." }, { "issue_ko": "방범창의 쇠창살 디자인(가로줄의 개수)이 레퍼런스 사진의 구조와 다릅니다.", "fix_en": "Change the window bars to match the reference photograph, featuring vertical bars with only a single horizontal bar in the middle." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Render the figures entirely as a reflection on the exterior of the window glass, rather than showing them inside the room.\n- Adjust the rooftop room door so it is only slightly open.\n- Modify the door's design, material, and color to exactly match the brown metal door in the location reference photograph.\n- Change the window bars to match the reference photograph, featuring vertical bars with only a single horizontal bar in the middle.\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)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02" }, "assigned": "L04B02", "choice": "Candidate B", "confident": true, "reason_ko": "샷 내용에 '창문에 비친 그림자'가 언급되어 옥탑방의 창문이 있는 공간이 필요합니다. 제공된 두 후보군 모두 창문이 포함된 옥상 외부의 동일한 서브공간을 보여주고 있으므로, 공간적 구분에 따른 명확한 변경 근거가 없어 현재 할당된 후보 B를 유지합니다.", "kept": "L04B02" } }, "S11sh3::variants": { "author_fp": "da516be6c07eb4c1", "author": { "variants": [ { "approach_ko": "눈높이의 타이트한 정면 클로즈업으로 표정과 코를 막은 손의 압박감을 강조한다.", "prompt_en": "Frame a tight, eye-level frontal close-up with her face and the hand covering her nose as the dominant visual information. Use a gently compressed perspective and very shallow focus, holding both eyes and the gripping fingers crisp while the stairway falls into soft nocturnal abstraction behind her. Keep the composition still, centered, and claustrophobic, with soft off-axis light shaping her youthful features without glamour." }, { "approach_ko": "가까운 광각과 비대칭 계단 구도로 인물의 정면 얼굴을 환경 속 불안한 순간으로 포착한다.", "prompt_en": "Photograph her from close range with a wider perspective, keeping her face fully frontal while allowing the stair geometry to cut diagonally through the frame and recede with pronounced depth. Place her slightly off-center in a waist-up composition, with the hand over her nose clearly readable and the surrounding stairway remaining spatially present. Use restrained handheld immediacy, deeper focus, and uneven motivated night illumination to make the moment feel abruptly observed rather than posed." } ] }, "reused": false }, "S11sh3": { "input_fingerprint": "df60742c8e29a8f2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 계단 중간에 서서 한 손으로 코를 막은 수리영의 정면 얼굴\n\nLOCATION (lock): A narrow residential alley and the courtyard of a low-rise multi-unit building, with a secluded bicycle parking spot and an exterior stairway leading to the rooftop apartment and its half-open metal door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains hastily parked in the rear of the courtyard, and her mobile phone remains 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 계단 중간에 서서 한 손으로 코를 막은 수리영의 정면 얼굴\n\nLOCATION (lock): A narrow residential alley and the courtyard of a low-rise multi-unit building, with a secluded bicycle parking spot and an exterior stairway leading to the rooftop apartment and its half-open metal door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains hastily parked in the rear of the courtyard, and her mobile phone remains 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight, eye-level frontal close-up with her face and the hand covering her nose as the dominant visual information. Use a gently compressed perspective and very shallow focus, holding both eyes and the gripping fingers crisp while the stairway falls into soft nocturnal abstraction behind her. Keep the composition still, centered, and claustrophobic, with soft off-axis light shaping her youthful features without glamour.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 계단 중간에 서서 한 손으로 코를 막은 수리영의 정면 얼굴\n\nLOCATION (lock): A narrow residential alley and the courtyard of a low-rise multi-unit building, with a secluded bicycle parking spot and an exterior stairway leading to the rooftop apartment and its half-open metal door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains hastily parked in the rear of the courtyard, and her mobile phone remains 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her from close range with a wider perspective, keeping her face fully frontal while allowing the stair geometry to cut diagonally through the frame and recede with pronounced depth. Place her slightly off-center in a waist-up composition, with the hand over her nose clearly readable and the surrounding stairway remaining spatially present. Use restrained handheld immediacy, deeper focus, and uneven motivated night illumination to make the moment feel abruptly observed rather than posed.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 계단 중간에 서서 한 손으로 코를 막은 수리영의 정면 얼굴\n\nLOCATION (lock): A narrow residential alley and the courtyard of a low-rise multi-unit building, with a secluded bicycle parking spot and an exterior stairway leading to the rooftop apartment and its half-open metal door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains hastily parked in the rear of the courtyard, and her mobile phone remains 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight, eye-level frontal close-up with her face and the hand covering her nose as the dominant visual information. Use a gently compressed perspective and very shallow focus, holding both eyes and the gripping fingers crisp while the stairway falls into soft nocturnal abstraction behind her. Keep the composition still, centered, and claustrophobic, with soft off-axis light shaping her youthful features without glamour.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 계단 중간에 서서 한 손으로 코를 막은 수리영의 정면 얼굴\n\nLOCATION (lock): A narrow residential alley and the courtyard of a low-rise multi-unit building, with a secluded bicycle parking spot and an exterior stairway leading to the rooftop apartment and its half-open metal door. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong’s bicycle remains hastily parked in the rear of the courtyard, and her mobile phone remains 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her from close range with a wider perspective, keeping her face fully frontal while allowing the stair geometry to cut diagonally through the frame and recede with pronounced depth. Place her slightly off-center in a waist-up composition, with the hand over her nose clearly readable and the surrounding stairway remaining spatially present. Use restrained handheld immediacy, deeper focus, and uneven motivated night illumination to make the moment feel abruptly observed rather than posed.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B02.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": "A", "score": 6, "verdict_ko": "'정면 얼굴' 구도는 우수하나 코를 명확히 막기보다는 입 주변을 덮고 있어 동작 묘사가 다소 아쉽습니다." }, { "label": "B", "score": 5, "verdict_ko": "요구된 '얼굴' 중심의 프레이밍보다 너무 넓게 촬영되었으며, 배경에 있어야 할 자전거가 누락되었습니다." }, { "label": "C", "score": 8, "verdict_ko": "지정된 '정면 얼굴' 구도와 코를 막은 동작을 완벽히 구현했으며, 배경의 자전거와 옥상 문까지 정확히 묘사했습니다." }, { "label": "D", "score": 4, "verdict_ko": "프레이밍이 너무 넓게 잡혔고, 빈 손으로 묘사되어 휴대폰 유지 상태를 위반했습니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "A", "B", "D" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "'정면 얼굴' 구도는 우수하나 코를 명확히 막기보다는 입 주변을 덮고 있어 동작 묘사가 다소 아쉽습니다." }, { "label": "B", "score": 5, "verdict_ko": "요구된 '얼굴' 중심의 프레이밍보다 너무 넓게 촬영되었으며, 배경에 있어야 할 자전거가 누락되었습니다." }, { "label": "C", "score": 8, "verdict_ko": "지정된 '정면 얼굴' 구도와 코를 막은 동작을 완벽히 구현했으며, 배경의 자전거와 옥상 문까지 정확히 묘사했습니다." }, { "label": "D", "score": 4, "verdict_ko": "프레이밍이 너무 넓게 잡혔고, 빈 손으로 묘사되어 휴대폰 유지 상태를 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "C", "A", "D" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "'정면 얼굴'에 부합하는 클로즈업 프레이밍과 코를 막는 동작을 가장 정확하게 연출했습니다." }, { "label": "C", "score": 6, "verdict_ko": "코를 막는 동작과 소품은 정확하나 '정면 얼굴'을 강조하기에는 프레이밍이 다소 넓습니다." }, { "label": "A", "score": 5, "verdict_ko": "동작과 자전거 배치는 적절하지만 샷이 너무 넓어 '정면 얼굴'의 의도를 벗어납니다." }, { "label": "D", "score": 4, "verdict_ko": "얼굴 중심의 프레이밍은 좋으나 코를 막는 대신 입을 가리고 있어 핵심 동작을 위반했습니다." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "B", "D", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "'정면 얼굴'에 부합하는 클로즈업 프레이밍과 코를 막는 동작을 가장 정확하게 연출했습니다." }, { "label": "B", "score": 6, "verdict_ko": "코를 막는 동작과 소품은 정확하나 '정면 얼굴'을 강조하기에는 프레이밍이 다소 넓습니다." }, { "label": "D", "score": 5, "verdict_ko": "동작과 자전거 배치는 적절하지만 샷이 너무 넓어 '정면 얼굴'의 의도를 벗어납니다." }, { "label": "A", "score": 4, "verdict_ko": "얼굴 중심의 프레이밍은 좋으나 코를 막는 대신 입을 가리고 있어 핵심 동작을 위반했습니다." } ] }, "combined": { "totals": { "A": 10, "B": 11, "C": 15, "D": 9 }, "ranking": [ "C", "B", "A", "D" ], "agreement": true, "policy": 1 } }, "totals": { "A": 10, "B": 11, "C": 15, "D": 9 }, "selected": "C", "ranking": [ "C", "B", "A", "D" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "'정면 얼굴' 구도는 우수하나 코를 명확히 막기보다는 입 주변을 덮고 있어 동작 묘사가 다소 아쉽습니다." }, { "label": "B", "score": 5, "verdict_ko": "요구된 '얼굴' 중심의 프레이밍보다 너무 넓게 촬영되었으며, 배경에 있어야 할 자전거가 누락되었습니다." }, { "label": "C", "score": 8, "verdict_ko": "지정된 '정면 얼굴' 구도와 코를 막은 동작을 완벽히 구현했으며, 배경의 자전거와 옥상 문까지 정확히 묘사했습니다." }, { "label": "D", "score": 4, "verdict_ko": "프레이밍이 너무 넓게 잡혔고, 빈 손으로 묘사되어 휴대폰 유지 상태를 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "캐릭터가 계단 중간이 아닌 계단이 시작되는 맨 아래 평지에 서 있습니다.", "fix_en": "Position the character so she is standing halfway up the flight of stairs instead of at the bottom." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Position the character so she is standing halfway up the flight of stairs instead of at the bottom.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L12B01", "B": "L12B02", "C": "L12B03", "D": "L12B04" }, "assigned": "L12B02", "choice": "Candidate B", "confident": true, "reason_ko": "\"계단 중간에 서서\"라는 지문을 통해 인물이 위치한 계단 공간이 직접적으로 보이는 Candidate B가 가장 적합합니다.", "kept": "L12B02" } }, "S11sh4::variants": { "author_fp": "5ea50141ee786191", "author": { "variants": [ { "approach_ko": "망원 느낌의 정면 대칭 구도로 반쯤 열린 철문을 고립시키는 정적인 시선.", "prompt_en": "Photograph from a restrained medium-long distance on a square, head-on axis to the entrance, with a compressed lens feel that flattens the stairway and rooftop structure into taut architectural layers. Hold the half-open door near the exact visual center, using the surrounding walls and openings as a rigid frame; keep the courtyard and parked bicycle low and subordinate in the composition. Preserve the location’s established night illumination with controlled contrast and a quiet, watchful stillness." }, { "approach_ko": "안마당을 넓게 담고 계단의 선으로 정면의 철문까지 시선을 끌어올리는 깊은 공간 구도.", "prompt_en": "Use a wider environmental view from a low courtyard height while keeping the entrance face-on, letting the exterior stairway form a strong ascending path toward the half-open door. Build pronounced foreground-to-background depth, with the parked bicycle anchoring the lower courtyard and the doorway held smaller but unmistakably dominant at the end of the composition. Render the established night light naturally, allowing darker architectural planes and open space to create suspense without obscuring structural detail." } ] }, "reused": false }, "S11sh4": { "input_fingerprint": "802e797dd592ddea", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 반쯤 열린 옥탑방 현관 철문이 정면으로 보이는 시점 쇼트\n\nLOCATION (lock): A narrow residential alley and the courtyard of a low-rise multi-unit building, with a secluded bicycle parking spot and an exterior stairway leading to the rooftop apartment and its half-open metal 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 iron entrance door remains half open; Su-ri-yeong’s bicycle stays parked below in the courtyard.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 반쯤 열린 옥탑방 현관 철문이 정면으로 보이는 시점 쇼트\n\nLOCATION (lock): A narrow residential alley and the courtyard of a low-rise multi-unit building, with a secluded bicycle parking spot and an exterior stairway leading to the rooftop apartment and its half-open metal 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 iron entrance door remains half open; Su-ri-yeong’s bicycle stays parked below in the courtyard.\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 restrained medium-long distance on a square, head-on axis to the entrance, with a compressed lens feel that flattens the stairway and rooftop structure into taut architectural layers. Hold the half-open door near the exact visual center, using the surrounding walls and openings as a rigid frame; keep the courtyard and parked bicycle low and subordinate in the composition. Preserve the location’s established night illumination with controlled contrast and a quiet, watchful 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 반쯤 열린 옥탑방 현관 철문이 정면으로 보이는 시점 쇼트\n\nLOCATION (lock): A narrow residential alley and the courtyard of a low-rise multi-unit building, with a secluded bicycle parking spot and an exterior stairway leading to the rooftop apartment and its half-open metal 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 iron entrance door remains half open; Su-ri-yeong’s bicycle stays parked below in the courtyard.\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 environmental view from a low courtyard height while keeping the entrance face-on, letting the exterior stairway form a strong ascending path toward the half-open door. Build pronounced foreground-to-background depth, with the parked bicycle anchoring the lower courtyard and the doorway held smaller but unmistakably dominant at the end of the composition. Render the established night light naturally, allowing darker architectural planes and open space to create suspense without obscuring structural 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": 4, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "옥탑방 철문이 반쯤 열린 모습과 자전거는 구현했으나, 지시된 정면 시점 쇼트를 따르지 않고 레퍼런스 사진의 원경 구도를 그대로 모방했습니다." }, { "label": "B", "score": 3, "verdict_ko": "지시된 정면 시점 쇼트를 무시하고 레퍼런스 구도를 모방했으며, 옥탑방 문이 반쯤 열린 철문의 형태가 아닌 단순한 빛으로 잘못 묘사되었습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B04.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/background_chain/seed_bg_rooftop_villa_complex_sel.png" } ], "critique": { "issues": [ { "issue_ko": "카메라 앵글이 옥탑방 현관문을 정면으로 향하지 않고, 레퍼런스 사진을 모방하여 측면 위에서 내려다보는 구도로 촬영되었습니다.", "fix_en": "Reposition the camera to a square, head-on axis directly facing the rooftop entrance door, holding the door near the exact visual center." }, { "issue_ko": "자전거가 건물 마당(courtyard) 안이 아닌 바깥 골목길에 주차되어 있습니다.", "fix_en": "Move the parked bicycle from the exterior alley into the walled courtyard below the stairway." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Reposition the camera to a square, head-on axis directly facing the rooftop entrance door, holding the door near the exact visual center.\n- Move the parked bicycle from the exterior alley into the walled courtyard below the stairway.\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)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S10sh6" }, "lane_policy": "ab_select_bypass:bg_only", "plate_select": { "candidates": { "A": "L12B01", "B": "L12B02", "C": "L12B03", "D": "L12B04" }, "assigned": "L12B04", "choice": "Candidate D", "confident": true, "reason_ko": "텍스트에 명시된 '반쯤 열린 옥탑방 현관 철문이 정면으로 보이는' 공간을 정확히 보여주는 이미지는 후보 D입니다.", "kept": "L12B04" } }, "S12sh6::variants": { "author_fp": "77b84f92939d99ce", "author": { "variants": [ { "approach_ko": "바닥 높이의 밀착된 정면 시선으로 떨군 얼굴과 찢긴 어깨·쇄골을 압박감 있게 포착한다.", "prompt_en": "Photograph her upper body in a tight, floor-level frontal close shot, with the dropped head dominating the upper frame and the torn, bleeding shoulder and collarbone held in stark visual emphasis. Keep the curtain close to the frame edge for a claustrophobic enclosure, using shallow focus and the uneven television-and-lamp glow to model her slumped weight and cold stillness." }, { "approach_ko": "높은 사선 시점에서 커튼 뒤에 무너진 상체와 지지면에 떨어진 손의 관계를 냉정하게 내려다본다.", "prompt_en": "Use a steep high-angle view from slightly to the side, looking down on her collapsed upper body behind the curtain. Build the composition from the bowed head through the injured shoulder to the fully supported clenched hand and crumpled photograph, with restrained depth and dim television and lamp light creating a detached, forensic stillness." } ] }, "reused": false }, "S12sh6": { "input_fingerprint": "e0a8c81b3753abda", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흘러내리는 끔찍한 상처를 입은 채 바닥에 앉아 고개를 떨군 민숙의 상체.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. 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): Min-suk is seated on the floor behind the curtain with her torso slumped and her head dropped forward; one cold, stiff hand remains clenched around a crumpled old 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’s corpse remains seated on the floor behind the curtain with her torso slumped and her head dropped forward, her shoulder and collarbone torn and bleeding. One cold, stiff hand remains clenched around a crumpled old photograph, and the circular mark remains on her 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: 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흘러내리는 끔찍한 상처를 입은 채 바닥에 앉아 고개를 떨군 민숙의 상체.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. 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): Min-suk is seated on the floor behind the curtain with her torso slumped and her head dropped forward; one cold, stiff hand remains clenched around a crumpled old 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’s corpse remains seated on the floor behind the curtain with her torso slumped and her head dropped forward, her shoulder and collarbone torn and bleeding. One cold, stiff hand remains clenched around a crumpled old photograph, and the circular mark remains on her 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: 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her upper body in a tight, floor-level frontal close shot, with the dropped head dominating the upper frame and the torn, bleeding shoulder and collarbone held in stark visual emphasis. Keep the curtain close to the frame edge for a claustrophobic enclosure, using shallow focus and the uneven television-and-lamp glow to model her slumped weight and cold 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흘러내리는 끔찍한 상처를 입은 채 바닥에 앉아 고개를 떨군 민숙의 상체.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. 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): Min-suk is seated on the floor behind the curtain with her torso slumped and her head dropped forward; one cold, stiff hand remains clenched around a crumpled old 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’s corpse remains seated on the floor behind the curtain with her torso slumped and her head dropped forward, her shoulder and collarbone torn and bleeding. One cold, stiff hand remains clenched around a crumpled old photograph, and the circular mark remains on her 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: 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a steep high-angle view from slightly to the side, looking down on her collapsed upper body behind the curtain. Build the composition from the bowed head through the injured shoulder to the fully supported clenched hand and crumpled photograph, with restrained depth and dim television and lamp light creating a detached, 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.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흘러내리는 끔찍한 상처를 입은 채 바닥에 앉아 고개를 떨군 민숙의 상체.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. 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): Min-suk is seated on the floor behind the curtain with her torso slumped and her head dropped forward; one cold, stiff hand remains clenched around a crumpled old 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’s corpse remains seated on the floor behind the curtain with her torso slumped and her head dropped forward, her shoulder and collarbone torn and bleeding. One cold, stiff hand remains clenched around a crumpled old photograph, and the circular mark remains on her 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: 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her upper body in a tight, floor-level frontal close shot, with the dropped head dominating the upper frame and the torn, bleeding shoulder and collarbone held in stark visual emphasis. Keep the curtain close to the frame edge for a claustrophobic enclosure, using shallow focus and the uneven television-and-lamp glow to model her slumped weight and cold 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 살점이 뜯겨 피가 흘러내리는 끔찍한 상처를 입은 채 바닥에 앉아 고개를 떨군 민숙의 상체.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. 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): Min-suk is seated on the floor behind the curtain with her torso slumped and her head dropped forward; one cold, stiff hand remains clenched around a crumpled old 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’s corpse remains seated on the floor behind the curtain with her torso slumped and her head dropped forward, her shoulder and collarbone torn and bleeding. One cold, stiff hand remains clenched around a crumpled old photograph, and the circular mark remains on her 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: 민숙 (한국인 여성, 40대 초반, 부드러운 중년 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a steep high-angle view from slightly to the side, looking down on her collapsed upper body behind the curtain. Build the composition from the bowed head through the injured shoulder to the fully supported clenched hand and crumpled photograph, with restrained depth and dim television and lamp light creating a detached, 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." }, "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S12sh6.png" }, { "label": "CHARACTER REFERENCE — 민숙: 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S12sh6.png" }, { "label": "CHARACTER REFERENCE — 민숙: 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.png" }, { "label": "CHARACTER REFERENCE — 민숙: 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.png" }, { "label": "CHARACTER REFERENCE — 민숙: 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": "지정된 공간과 시체 자세, 무중력 상태의 늘어진 손(중력 법칙)을 가장 정확히 구현함." }, { "label": "B", "score": 5, "verdict_ko": "배경과 의상은 정확하나, 사진을 든 손이 허공에 떠 있어 무의식/사후 중력 지침을 위반함." }, { "label": "C", "score": 3, "verdict_ko": "의상이 참조와 다르고, 소품을 쥐고 있지 않으며 실내 건축 구조가 심하게 변형됨." }, { "label": "D", "score": 4, "verdict_ko": "손의 자세는 중력을 잘 따랐으나, 참조된 실내 건축(슬라이딩 도어 등)과 완전히 불일치함." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 공간과 시체 자세, 무중력 상태의 늘어진 손(중력 법칙)을 가장 정확히 구현함." }, { "label": "B", "score": 5, "verdict_ko": "배경과 의상은 정확하나, 사진을 든 손이 허공에 떠 있어 무의식/사후 중력 지침을 위반함." }, { "label": "C", "score": 3, "verdict_ko": "의상이 참조와 다르고, 소품을 쥐고 있지 않으며 실내 건축 구조가 심하게 변형됨." }, { "label": "D", "score": 4, "verdict_ko": "손의 자세는 중력을 잘 따랐으나, 참조된 실내 건축(슬라이딩 도어 등)과 완전히 불일치함." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "C", "B", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "장소, 의상, 중력 지침에 순응하는 자세를 완벽히 재현했으나, 요구된 끔찍한 상처 묘사와 타이트한 상체 프레이밍이 누락되었습니다." }, { "label": "C", "score": 5, "verdict_ko": "장소와 의상은 일치하지만, 바닥이 아닌 곳에 앉아 있고 사진을 든 손이 허공에 떠 있어 중력 지침을 위반했습니다." }, { "label": "B", "score": 4, "verdict_ko": "바닥에 앉은 자세는 맞으나 기준 사진과 다른 패턴 스웨터를 입었고 방의 구조가 지정된 장소와 어긋납니다." }, { "label": "A", "score": 3, "verdict_ko": "지정된 장소의 구조와 완전히 다르고 프레이밍이 너무 넓으며, 상처 묘사도 전혀 없습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "장소, 의상, 중력 지침에 순응하는 자세를 완벽히 재현했으나, 요구된 끔찍한 상처 묘사와 타이트한 상체 프레이밍이 누락되었습니다." }, { "label": "B", "score": 5, "verdict_ko": "장소와 의상은 일치하지만, 바닥이 아닌 곳에 앉아 있고 사진을 든 손이 허공에 떠 있어 중력 지침을 위반했습니다." }, { "label": "C", "score": 4, "verdict_ko": "바닥에 앉은 자세는 맞으나 기준 사진과 다른 패턴 스웨터를 입었고 방의 구조가 지정된 장소와 어긋납니다." }, { "label": "D", "score": 3, "verdict_ko": "지정된 장소의 구조와 완전히 다르고 프레이밍이 너무 넓으며, 상처 묘사도 전혀 없습니다." } ] }, "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 공간과 시체 자세, 무중력 상태의 늘어진 손(중력 법칙)을 가장 정확히 구현함." }, { "label": "B", "score": 5, "verdict_ko": "배경과 의상은 정확하나, 사진을 든 손이 허공에 떠 있어 무의식/사후 중력 지침을 위반함." }, { "label": "C", "score": 3, "verdict_ko": "의상이 참조와 다르고, 소품을 쥐고 있지 않으며 실내 건축 구조가 심하게 변형됨." }, { "label": "D", "score": 4, "verdict_ko": "손의 자세는 중력을 잘 따랐으나, 참조된 실내 건축(슬라이딩 도어 등)과 완전히 불일치함." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.png" }, { "label": "CHARACTER REFERENCE — 민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "상체만 집중적으로 보여주는 타이트한 클로즈업 샷이어야 하나, 인물의 전신과 주변 공간이 넓게 포함된 와이드 샷으로 연출되었습니다.", "fix_en": "Change the camera framing to a tight, floor-level close-up shot focused exclusively on the character's upper body." }, { "issue_ko": "어깨와 쇄골에 피가 흐르는 찢겨진 상처가 묘사되어야 하지만, 훼손되지 않은 스웨터를 입고 상처가 전혀 보이지 않습니다.", "fix_en": "Modify the clothing and skin at the shoulder and collarbone to reveal a gruesome, bleeding wound with torn flesh." }, { "issue_ko": "손에 낡고 구겨진 사진을 쥐고 있어야 하나, 접힌 흰색 휴지나 종이 조각을 들고 있습니다.", "fix_en": "Replace the folded white paper item in the character's left hand with a crumpled, old photograph." }, { "issue_ko": "시신이므로 근육의 힘이 완전히 빠져 있어야 하나, 바닥을 짚은 오른손에 체중을 지탱하는 힘이 들어가 있습니다.", "fix_en": "Adjust the right hand resting on the floor so it lies completely flat and limp, showing zero muscular tension." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the camera framing to a tight, floor-level close-up shot focused exclusively on the character's upper body.\n- Modify the clothing and skin at the shoulder and collarbone to reveal a gruesome, bleeding wound with torn flesh.\n- Replace the folded white paper item in the character's left hand with a crumpled, old photograph.\n- Adjust the right hand resting on the floor so it lies completely flat and limp, showing zero muscular tension.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B06", "choice": "Candidate F", "confident": false, "reason_ko": "\"바닥에 앉아\"라는 묘사 외에 특정 공간을 지시하는 단서가 없어, 기존에 할당된 컷을 유지합니다.", "kept": "L05B06" } }, "S12sh9::variants": { "author_fp": "494622e576c8e1f3", "author": { "variants": [ { "approach_ko": "손과 사진만 프레임을 채우는 정면에 가까운 극단적 클로즈업으로 굳은 촉감과 구김을 강조한다.", "prompt_en": "Photograph an extreme close-up from nearly perpendicular to the clenched hand, filling the frame with the rigid fingers and crumpled old photograph while excluding the rest of the body. Use shallow focus that holds the gripping fingertips and crushed folds sharply, with soft television and lamp light revealing the cold, dry texture and leaving the frame edges subdued." }, { "approach_ko": "바닥 높이의 비스듬한 측면 시점과 스치는 빛으로 손가락의 경직과 사진의 찌그러진 윤곽을 입체적으로 드러낸다.", "prompt_en": "Shoot from floor height at a tight oblique side angle, letting the clenched hand form a stark foreground profile around the compressed photograph. Use raking mixed television and lamp light to catch the knuckles, fixed finger joints, and raised paper creases, with a narrow plane of focus and the apartment interior dissolving into dim, indistinct depth." } ] }, "reused": false }, "S12sh9": { "input_fingerprint": "53aa8834a02a5b8f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 민숙의 차갑게 굳은 손안에 구겨진 채 쥐어져 있는 낡은 사진 한 장의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-suk remains dead, seated on the floor behind the curtain with her torso slumped and her head dropped forward, with the torn shoulder wounds, while the crumpled old photograph stays clenched in her cold hand and the wrist mark remains present.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 민숙의 차갑게 굳은 손안에 구겨진 채 쥐어져 있는 낡은 사진 한 장의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-suk remains dead, seated on the floor behind the curtain with her torso slumped and her head dropped forward, with the torn shoulder wounds, while the crumpled old photograph stays clenched in her cold hand and the wrist mark remains present.\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 an extreme close-up from nearly perpendicular to the clenched hand, filling the frame with the rigid fingers and crumpled old photograph while excluding the rest of the body. Use shallow focus that holds the gripping fingertips and crushed folds sharply, with soft television and lamp light revealing the cold, dry texture and leaving the frame edges subdued.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 민숙의 차갑게 굳은 손안에 구겨진 채 쥐어져 있는 낡은 사진 한 장의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-suk remains dead, seated on the floor behind the curtain with her torso slumped and her head dropped forward, with the torn shoulder wounds, while the crumpled old photograph stays clenched in her cold hand and the wrist mark remains present.\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):\nShoot from floor height at a tight oblique side angle, letting the clenched hand form a stark foreground profile around the compressed photograph. Use raking mixed television and lamp light to catch the knuckles, fixed finger joints, and raised paper creases, with a narrow plane of focus and the apartment interior dissolving into dim, indistinct 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." }, "totals": { "A": 7, "B": 5 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "사진과 굳은 손에 오롯이 집중한 완벽한 클로즈업 앵글을 구현했으며, 배경의 TV와 조명 등 모든 지시 사항을 충실히 반영했습니다." }, { "label": "B", "score": 5, "verdict_ko": "공간의 맥락(커튼 등)을 보여주기 위해 프레임을 넓히는 바람에 샷 텍스트가 요구한 '사진의 클로즈업'이라는 핵심에서 다소 멀어졌습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.png" } ], "critique": { "issues": [ { "issue_ko": "구겨진 사진의 왼쪽 하단에 프롬프트에서 금지한 텍스트가 적혀 있습니다.", "fix_en": "Remove the black text from the bottom left corner of the crumpled photograph." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the black text from the bottom left corner of the crumpled 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": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S12sh6" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B06", "choice": "Candidate F", "confident": true, "reason_ko": "클로즈업 샷으로 손과 사진만 묘사되어 특정 공간(침실이나 거실 등)을 지시하는 단서가 없어, 기존에 할당된 컷을 그대로 유지합니다.", "kept": "L05B06" } }, "S12sh12::variants": { "author_fp": "41b87d0efe9bf398", "author": { "variants": [ { "approach_ko": "침대와 벽을 정면으로 바라보는 고정된 중거리 대칭 구도로 붉은 원형 표식을 중심에 선명하게 부각한다.", "prompt_en": "Use a locked-off, eye-level frontal composition with the bed forming a quiet horizontal base and the wall dominating the frame. Center the red circular mark above it in crisp focus, using the dim television and lamp illumination to create restrained tonal contrast and an unnervingly static, evidentiary stillness." }, { "approach_ko": "침대 가장자리 가까이에서 비스듬히 올려다보는 압축된 구도로 붉은 표식의 질감과 위압감을 강조한다.", "prompt_en": "Photograph from close to the bed at a low, oblique angle, letting the near edge of the bed anchor the foreground while the wall rises steeply behind it. Frame tightly around the red circular mark and its immediate surroundings, with shallow depth separation and raking dim interior light revealing its irregular, blood-like surface against the wall." } ] }, "reused": false }, "S12sh12": { "input_fingerprint": "6ff1139874b58af4", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 침대 위 벽면에 피로 그린 듯한 붉은색 원형 표식이 뚜렷하게 떠오른 정지 구도.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 the same wounds, but Su-ri-yeong has now removed the photograph from her hand and the wrist mark has turned red. A newly formed red circle remains visible 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 침대 위 벽면에 피로 그린 듯한 붉은색 원형 표식이 뚜렷하게 떠오른 정지 구도.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 the same wounds, but Su-ri-yeong has now removed the photograph from her hand and the wrist mark has turned red. A newly formed red circle remains visible 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):\nUse a locked-off, eye-level frontal composition with the bed forming a quiet horizontal base and the wall dominating the frame. Center the red circular mark above it in crisp focus, using the dim television and lamp illumination to create restrained tonal contrast and an unnervingly static, evidentiary 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, dim interior with television and lamp light.\n\nSHOT TEXT (authoritative, Korean): 침대 위 벽면에 피로 그린 듯한 붉은색 원형 표식이 뚜렷하게 떠오른 정지 구도.\n\nLOCATION (lock): A compact rooftop apartment with a living room, dining table, small kitchen, main bedroom, and a separate young woman's bedroom with a bed, curtains, and a window. The living room contains a cooled cooking pot, a television left on, and a dim floor or table lamp. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 the same wounds, but Su-ri-yeong has now removed the photograph from her hand and the wrist mark has turned red. A newly formed red circle remains visible 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 from close to the bed at a low, oblique angle, letting the near edge of the bed anchor the foreground while the wall rises steeply behind it. Frame tightly around the red circular mark and its immediate surroundings, with shallow depth separation and raking dim interior light revealing its irregular, blood-like surface against the wall.\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": "레퍼런스 이미지의 공간 구조(좌측 거실 문, 우측 창문)와 조명 설정을 완벽하게 재현하며 프롬프트의 지시사항을 잘 따랐습니다." }, { "label": "B", "score": 3, "verdict_ko": "벽면의 붉은 표식은 표현되었으나, 방의 구조가 레퍼런스와 전혀 일치하지 않아 로케이션 제약을 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.png" } ], "critique": { "issues": [ { "issue_ko": "붉은 원형 표식이 침대 위가 아닌 침대 왼쪽의 빈 벽면에 위치해 있습니다.", "fix_en": "Move the red circular mark to the right so it is on the wall directly above the bed." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the red circular mark to the right so it is on the wall directly above the bed.\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)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S12sh9" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B06", "choice": "Candidate F", "confident": true, "reason_ko": "\"침대 위 벽면\"이라는 묘사에 따라 침대와 그 위쪽 벽면이 온전히 드러나는 침실 공간이 필요하며, 기존에 할당된 후보가 이를 충족하므로 유지합니다.", "kept": "L05B06" } }, "S13sh2::variants": { "author_fp": "58f63c79901524fb", "author": { "variants": [ { "approach_ko": "눈높이의 밀착된 정면 클로즈업으로 멍한 표정과 붉은 경광등의 순간적인 얼굴 조명을 강조한다.", "prompt_en": "Photograph her in a tight, eye-level chest-up close-up, nearly frontal and held with unnerving stillness. Use shallow focus to isolate her youthful face and unfixed gaze, with the police beacon’s red light raking across one side of her face and clothing while the rooftop exterior dissolves into subdued night blur." }, { "approach_ko": "비스듬한 높은 시점의 환경형 상반신 숏으로 인물의 고립감과 계단·옥상 출입구의 깊이를 부각한다.", "prompt_en": "Frame an oblique upper-body portrait from a distinctly elevated viewpoint, keeping her smaller and off-center within the hard descending geometry of the rooftop exit and steep stairs. Preserve more environmental depth and texture than in an intimate close-up; let the red emergency light spread unevenly across her and the surrounding surfaces, making her rigid, vacant stillness feel exposed within the night exterior." } ] }, "reused": false }, "S13sh2": { "input_fingerprint": "21a065907c2d037f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 경찰차의 붉은 경광등 불빛을 받으며 초점을 잃고 멍하니 서 있는 수리영의 상체.\n\nLOCATION (lock): The exterior of a rooftop apartment building, including the rooftop exit, steep connecting stairs, lower courtyard, and nearby vehicle access where responders arrive. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still has the old photograph she removed from Minsuk’s hand. Upstairs, Minsuk’s seated corpse, the blood traces, and the red wall circle remain until the subsequent unexplained cleanup.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 경찰차의 붉은 경광등 불빛을 받으며 초점을 잃고 멍하니 서 있는 수리영의 상체.\n\nLOCATION (lock): The exterior of a rooftop apartment building, including the rooftop exit, steep connecting stairs, lower courtyard, and nearby vehicle access where responders arrive. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still has the old photograph she removed from Minsuk’s hand. Upstairs, Minsuk’s seated corpse, the blood traces, and the red wall circle remain until the subsequent unexplained cleanup.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a tight, eye-level chest-up close-up, nearly frontal and held with unnerving stillness. Use shallow focus to isolate her youthful face and unfixed gaze, with the police beacon’s red light raking across one side of her face and clothing while the rooftop exterior dissolves into subdued night 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 경찰차의 붉은 경광등 불빛을 받으며 초점을 잃고 멍하니 서 있는 수리영의 상체.\n\nLOCATION (lock): The exterior of a rooftop apartment building, including the rooftop exit, steep connecting stairs, lower courtyard, and nearby vehicle access where responders arrive. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still has the old photograph she removed from Minsuk’s hand. Upstairs, Minsuk’s seated corpse, the blood traces, and the red wall circle remain until the subsequent unexplained cleanup.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame an oblique upper-body portrait from a distinctly elevated viewpoint, keeping her smaller and off-center within the hard descending geometry of the rooftop exit and steep stairs. Preserve more environmental depth and texture than in an intimate close-up; let the red emergency light spread unevenly across her and the surrounding surfaces, making her rigid, vacant stillness feel exposed within the night 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.", "C": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 경찰차의 붉은 경광등 불빛을 받으며 초점을 잃고 멍하니 서 있는 수리영의 상체.\n\nLOCATION (lock): The exterior of a rooftop apartment building, including the rooftop exit, steep connecting stairs, lower courtyard, and nearby vehicle access where responders arrive. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still has the old photograph she removed from Minsuk’s hand. Upstairs, Minsuk’s seated corpse, the blood traces, and the red wall circle remain until the subsequent unexplained cleanup.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a tight, eye-level chest-up close-up, nearly frontal and held with unnerving stillness. Use shallow focus to isolate her youthful face and unfixed gaze, with the police beacon’s red light raking across one side of her face and clothing while the rooftop exterior dissolves into subdued night 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.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 경찰차의 붉은 경광등 불빛을 받으며 초점을 잃고 멍하니 서 있는 수리영의 상체.\n\nLOCATION (lock): The exterior of a rooftop apartment building, including the rooftop exit, steep connecting stairs, lower courtyard, and nearby vehicle access where responders arrive. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still has the old photograph she removed from Minsuk’s hand. Upstairs, Minsuk’s seated corpse, the blood traces, and the red wall circle remain until the subsequent unexplained cleanup.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame an oblique upper-body portrait from a distinctly elevated viewpoint, keeping her smaller and off-center within the hard descending geometry of the rooftop exit and steep stairs. Preserve more environmental depth and texture than in an intimate close-up; let the red emergency light spread unevenly across her and the surrounding surfaces, making her rigid, vacant stillness feel exposed within the night 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." }, "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S13sh2.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S13sh2.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B03.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", "B", "D", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 '상체' 프레이밍을 정확히 구현했으며, 사진 소품의 방향과 지정된 배경 및 조명 조건을 모두 완벽하게 충족합니다." }, { "label": "B", "score": 5, "verdict_ko": "배경과 조명은 기준에 부합하나, 요구된 '상체' 샷 대신 전신 풀샷으로 연출되어 프레이밍 우선순위를 위반했습니다." }, { "label": "D", "score": 4, "verdict_ko": "요구된 프레이밍보다 넓게 촬영되었으며, 배경 건축물에 유리문이 추가되는 등 지정된 로케이션을 훼손했습니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 언급되지 않은 시체를 계단에 임의로 추가하여 '등장인물 임의 추가 불가' 규정을 심각하게 위반했습니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "B", "D", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 '상체' 프레이밍을 정확히 구현했으며, 사진 소품의 방향과 지정된 배경 및 조명 조건을 모두 완벽하게 충족합니다." }, { "label": "B", "score": 5, "verdict_ko": "배경과 조명은 기준에 부합하나, 요구된 '상체' 샷 대신 전신 풀샷으로 연출되어 프레이밍 우선순위를 위반했습니다." }, { "label": "D", "score": 4, "verdict_ko": "요구된 프레이밍보다 넓게 촬영되었으며, 배경 건축물에 유리문이 추가되는 등 지정된 로케이션을 훼손했습니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 언급되지 않은 시체를 계단에 임의로 추가하여 '등장인물 임의 추가 불가' 규정을 심각하게 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A", "C", "D" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "'상체' 샷 규격, 붉은 경광등 조명, 공허한 표정, 자연스러운 소품 방향을 모두 정확히 준수했습니다." }, { "label": "A", "score": 5, "verdict_ko": "상체 샷보다 화각이 넓으며, 우측 배경 계단에 정체불명의 팔이 생성된 심각한 오류가 있습니다." }, { "label": "C", "score": 4, "verdict_ko": "상체 샷 지시를 완전히 무시하고, 금지된 레퍼런스 이미지의 카메라 구도를 그대로 복사했습니다." }, { "label": "D", "score": 3, "verdict_ko": "위층에 있어야 할 시신이 무단으로 배경에 등장하며, 소품 사진의 앞면을 카메라로 향하게 한 치명적 오류가 있습니다." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "D", "B", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "'상체' 샷 규격, 붉은 경광등 조명, 공허한 표정, 자연스러운 소품 방향을 모두 정확히 준수했습니다." }, { "label": "D", "score": 5, "verdict_ko": "상체 샷보다 화각이 넓으며, 우측 배경 계단에 정체불명의 팔이 생성된 심각한 오류가 있습니다." }, { "label": "B", "score": 4, "verdict_ko": "상체 샷 지시를 완전히 무시하고, 금지된 레퍼런스 이미지의 카메라 구도를 그대로 복사했습니다." }, { "label": "A", "score": 3, "verdict_ko": "위층에 있어야 할 시신이 무단으로 배경에 등장하며, 소품 사진의 앞면을 카메라로 향하게 한 치명적 오류가 있습니다." } ] }, "combined": { "totals": { "A": 6, "B": 9, "C": 14, "D": 9 }, "ranking": [ "C", "B", "D", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 9, "C": 14, "D": 9 }, "selected": "C", "ranking": [ "C", "B", "D", "A" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지시된 '상체' 프레이밍을 정확히 구현했으며, 사진 소품의 방향과 지정된 배경 및 조명 조건을 모두 완벽하게 충족합니다." }, { "label": "B", "score": 5, "verdict_ko": "배경과 조명은 기준에 부합하나, 요구된 '상체' 샷 대신 전신 풀샷으로 연출되어 프레이밍 우선순위를 위반했습니다." }, { "label": "D", "score": 4, "verdict_ko": "요구된 프레이밍보다 넓게 촬영되었으며, 배경 건축물에 유리문이 추가되는 등 지정된 로케이션을 훼손했습니다." }, { "label": "A", "score": 3, "verdict_ko": "샷 텍스트에 언급되지 않은 시체를 계단에 임의로 추가하여 '등장인물 임의 추가 불가' 규정을 심각하게 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B03.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "화면 좌측 하단에 보이는 조명이 경찰차의 경광등이 아닌 공사 현장용 둥근 덮개형 경고등으로 묘사되었습니다.", "fix_en": "Change the domed warning light on the bottom left into a standard modern police car lightbar." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the domed warning light on the bottom left into a standard modern police car lightbar.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L12B01", "B": "L12B02", "C": "L12B03", "D": "L12B04" }, "assigned": "L12B03", "choice": "Candidate C", "confident": true, "reason_ko": "\"경찰차의 붉은 경광등 불빛\"이라는 묘사에 부합하도록 경찰차가 도착해 있는 하단 마당/주차 구역이 적합합니다.", "kept": "L12B03" } }, "S13sh5::variants": { "author_fp": "2429030ee019f7e9", "author": { "variants": [ { "approach_ko": "정면 눈높이의 정적인 망원 클로즈업으로 충격에 굳은 표정을 대칭적으로 고립시킨다.", "prompt_en": "Photograph her in a locked, eye-level frontal close-up with a restrained telephoto feel, her face centered and nearly filling the frame. Keep both eyes and the parted lips critically sharp while the surroundings dissolve into darkness; let the red police-light illumination rake softly across one side of her face, leaving the opposite side in controlled shadow." }, { "approach_ko": "얼굴에 바짝 붙은 비스듬한 로우앵글로 불안정한 원근과 비대칭 구도를 강조한다.", "prompt_en": "Move extremely close from a slightly low, three-quarter angle with a subtly wide lens feel, creating uneasy perspective without distorting her anatomy. Frame her face asymmetrically with the nearer eye dominating the composition, keeping both widened eyes and the half-open mouth legible as red police light grazes the facial planes against a heavily defocused night background." } ] }, "reused": false }, "S13sh5": { "input_fingerprint": "b117302689debb2c", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜨고 입을 반쯤 벌린 채 충격으로 멈춘 혜수의 얼굴 클로즈업.\n\nLOCATION (lock): The exterior of a rooftop apartment building, including the rooftop exit, steep connecting stairs, lower courtyard, and nearby vehicle access where responders arrive. 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 nighttime courtyard look, red police-light illumination, and the ongoing emergency atmosphere from the reference. Exclude the stunned young woman and the wider police-car composition, keeping only the investigator's shocked face in 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): The wounded corpse, blood traces, and red wall circle are still upstairs as Hye-su reacts and heads toward the rooftop room; Su-ri-yeong retains the old photograph unless the later cleanup removes it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜨고 입을 반쯤 벌린 채 충격으로 멈춘 혜수의 얼굴 클로즈업.\n\nLOCATION (lock): The exterior of a rooftop apartment building, including the rooftop exit, steep connecting stairs, lower courtyard, and nearby vehicle access where responders arrive. 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 nighttime courtyard look, red police-light illumination, and the ongoing emergency atmosphere from the reference. Exclude the stunned young woman and the wider police-car composition, keeping only the investigator's shocked face in 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): The wounded corpse, blood traces, and red wall circle are still upstairs as Hye-su reacts and heads toward the rooftop room; Su-ri-yeong retains the old photograph unless the later cleanup removes it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a locked, eye-level frontal close-up with a restrained telephoto feel, her face centered and nearly filling the frame. Keep both eyes and the parted lips critically sharp while the surroundings dissolve into darkness; let the red police-light illumination rake softly across one side of her face, leaving the opposite side in controlled 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜨고 입을 반쯤 벌린 채 충격으로 멈춘 혜수의 얼굴 클로즈업.\n\nLOCATION (lock): The exterior of a rooftop apartment building, including the rooftop exit, steep connecting stairs, lower courtyard, and nearby vehicle access where responders arrive. 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 nighttime courtyard look, red police-light illumination, and the ongoing emergency atmosphere from the reference. Exclude the stunned young woman and the wider police-car composition, keeping only the investigator's shocked face in 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): The wounded corpse, blood traces, and red wall circle are still upstairs as Hye-su reacts and heads toward the rooftop room; Su-ri-yeong retains the old photograph unless the later cleanup removes it.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nMove extremely close from a slightly low, three-quarter angle with a subtly wide lens feel, creating uneasy perspective without distorting her anatomy. Frame her face asymmetrically with the nearer eye dominating the composition, keeping both widened eyes and the half-open mouth legible as red police light grazes the facial planes against a heavily defocused night 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": 8, "B": 6 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 클로즈업 구도를 정확히 구현하고 경찰차를 프레임에서 배제하여 프롬프트의 의도를 잘 충족했습니다." }, { "label": "B", "score": 6, "verdict_ko": "충격받은 표정 묘사는 우수하나, 제외하도록 지시된 경찰차 일부가 프레임에 남아있어 감점 요소가 됩니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S13sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "캐릭터 레퍼런스에 있는 버건디색 터틀넥이 누락되어 가죽 재킷 아래로 맨목과 윗가슴이 노출되어 있습니다.", "fix_en": "Add the high-collared burgundy turtleneck underneath the brown leather jacket so no bare skin is visible on her neck and chest." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the high-collared burgundy turtleneck underneath the brown leather jacket so no bare skin is visible on her neck and chest.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S13sh2" } }, "S14sh3::variants": { "author_fp": "e7beb06ba6cdcc4d", "author": { "variants": [ { "approach_ko": "출입구 높이의 정면 와이드 숏으로 침대와 벽면의 지나친 질서와 비어 있는 공간을 냉정하게 강조한다.", "prompt_en": "Photograph from the entrance at standing eye level in a locked-off, frontal wide composition. Use restrained wide-angle perspective and deep focus, arranging the room’s existing elements in clean, rigid layers while giving the bed and wall strong geometric prominence. Flat, cool interior illumination with minimal shadow makes every surface clinically legible and turns the excessive order into the source of unease." }, { "approach_ko": "침대 가장자리 가까이의 낮은 사선 시점과 압축된 원근으로 매끈한 침구에서 벽면까지 불길한 시선을 유도한다.", "prompt_en": "Place the camera low and close beside the bed, looking obliquely across its immaculate surface toward the wall with a compressed, intimate lens feel. Let the foreground bedding occupy much of the frame while the wall and remaining established objects stack tightly behind it in controlled depth. Soft light raking from one side reveals minute texture without disturbing the pristine appearance, creating tension through stillness and negative space." } ] }, "reused": false }, "S14sh3": { "input_fingerprint": "45f58236b371cc98", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핏자국이나 붉은 표식 없이 지나치게 깨끗하게 정돈된 침대와 벽면의 정경.\n\nLOCATION (lock): A small rooftop apartment interior with an entrance, living and dining area, and a meticulously arranged bedroom containing a bed, curtain, backpack, and doll. Every surface appears unusually clean and undisturbed. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 is now unnaturally clean: the blood, red wall circle, and Minsuk’s corpse have all vanished, ending their prior continuity.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핏자국이나 붉은 표식 없이 지나치게 깨끗하게 정돈된 침대와 벽면의 정경.\n\nLOCATION (lock): A small rooftop apartment interior with an entrance, living and dining area, and a meticulously arranged bedroom containing a bed, curtain, backpack, and doll. Every surface appears unusually clean and undisturbed. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 is now unnaturally clean: the blood, red wall circle, and Minsuk’s corpse have all vanished, ending their prior continuity.\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 the entrance at standing eye level in a locked-off, frontal wide composition. Use restrained wide-angle perspective and deep focus, arranging the room’s existing elements in clean, rigid layers while giving the bed and wall strong geometric prominence. Flat, cool interior illumination with minimal shadow makes every surface clinically legible and turns the excessive order into the source of unease.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핏자국이나 붉은 표식 없이 지나치게 깨끗하게 정돈된 침대와 벽면의 정경.\n\nLOCATION (lock): A small rooftop apartment interior with an entrance, living and dining area, and a meticulously arranged bedroom containing a bed, curtain, backpack, and doll. Every surface appears unusually clean and undisturbed. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 is now unnaturally clean: the blood, red wall circle, and Minsuk’s corpse have all vanished, ending their prior continuity.\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 across its immaculate surface toward the wall with a compressed, intimate lens feel. Let the foreground bedding occupy much of the frame while the wall and remaining established objects stack tightly behind it in controlled depth. Soft light raking from one side reveals minute texture without disturbing the pristine appearance, creating tension through stillness and negative space.\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": 4 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "'침대와 벽면의 정경'이라는 지문에 맞춰 구도를 집중시켰으며, 야간 시간대와 요구된 소품들을 지나치게 깨끗한 상태로 잘 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "레퍼런스 이미지의 카메라 구도를 그대로 복사하지 말라는 지침을 어겼으며, 지문이 요구한 범위를 넘어 불필요하게 넓은 공간을 보여줍니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.png" } ], "critique": { "issues": [] }, "fix_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B07", "choice": "Candidate G", "confident": true, "reason_ko": "지나치게 깨끗하게 정돈된 침대와 벽면이 보여야 하므로, 구겨진 침대(C)가 아닌 단정하게 정리된 침대가 있는 현재 할당된 후보가 가장 적합합니다.", "kept": "L05B07" } }, "S14sh4::variants": { "author_fp": "35da45597ebe5560", "author": { "variants": [ { "approach_ko": "눈높이 정면 밀착 클로즈업으로 굳은 표정과 커튼을 향한 시선을 포착한다.", "prompt_en": "Frame an extremely tight eye-level close-up, holding her face almost frontally while her gaze pulls decisively toward the curtain side of the frame. Use shallow focus to isolate the minute rigidity around her eyes, lips, and jaw, with the spotless bedroom and curtain reduced to a soft, recognizable background; subdued cool nighttime light gives her fear-stricken pallor a realistic cast without altering her natural skin or eyes." }, { "approach_ko": "커튼 가까이에서 바라본 측면 클로즈업으로 시선의 대상과 얼어붙은 반응을 한 화면에 압축한다.", "prompt_en": "Place the camera close to the curtain area and photograph her in a compressed profile close-up, with a softly blurred edge of the curtain in the near foreground and her frozen face in crisp focus beyond it. Keep the camera slightly below eye level so her fixed gaze carries tension across the narrow depth of the frame; let gentle side light skim her cheek and catch natural moisture in her eyes while the rest of the immaculate room falls into subdued darkness." } ] }, "reused": false }, "S14sh4": { "input_fingerprint": "c3e024f7fb9587a2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 창가 커튼 쪽을 보며 얼굴이 파랗게 질린 채 굳어버린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): A small rooftop apartment interior with an entrance, living and dining area, and a meticulously arranged bedroom containing a bed, curtain, backpack, and doll. Every surface appears unusually clean and undisturbed. 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 spotless bedroom state, curtain placement, bed area, and subdued nighttime light from the reference. Exclude the wide empty-room composition and any blood, corpse, or red symbol, using the setting only behind the frightened 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 curtain area remains empty, and the formerly bloody room stays excessively clean with no corpse or red mark visible.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 창가 커튼 쪽을 보며 얼굴이 파랗게 질린 채 굳어버린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): A small rooftop apartment interior with an entrance, living and dining area, and a meticulously arranged bedroom containing a bed, curtain, backpack, and doll. Every surface appears unusually clean and undisturbed. 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 spotless bedroom state, curtain placement, bed area, and subdued nighttime light from the reference. Exclude the wide empty-room composition and any blood, corpse, or red symbol, using the setting only behind the frightened 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 curtain area remains empty, and the formerly bloody room stays excessively clean with no corpse or red mark visible.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame an extremely tight eye-level close-up, holding her face almost frontally while her gaze pulls decisively toward the curtain side of the frame. Use shallow focus to isolate the minute rigidity around her eyes, lips, and jaw, with the spotless bedroom and curtain reduced to a soft, recognizable background; subdued cool nighttime light gives her fear-stricken pallor a realistic cast without altering her natural skin or eyes.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 창가 커튼 쪽을 보며 얼굴이 파랗게 질린 채 굳어버린 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): A small rooftop apartment interior with an entrance, living and dining area, and a meticulously arranged bedroom containing a bed, curtain, backpack, and doll. Every surface appears unusually clean and undisturbed. 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 spotless bedroom state, curtain placement, bed area, and subdued nighttime light from the reference. Exclude the wide empty-room composition and any blood, corpse, or red symbol, using the setting only behind the frightened 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 curtain area remains empty, and the formerly bloody room stays excessively clean with no corpse or red mark visible.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera close to the curtain area and photograph her in a compressed profile close-up, with a softly blurred edge of the curtain in the near foreground and her frozen face in crisp focus beyond it. Keep the camera slightly below eye level so her fixed gaze carries tension across the narrow depth of the frame; let gentle side light skim her cheek and catch natural moisture in her eyes while the rest of the immaculate room falls into subdued darkness.\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": 4 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "창가 커튼을 향한 시선과 굳은 표정은 물론, 배경의 침대와 소품 배치까지 프롬프트와 레퍼런스를 정확히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "창문을 등지고 있어 '창가 커튼 쪽을 보며'라는 핵심적인 시선 방향 지시를 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S14sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물의 얼굴에 혈색이 돌고 따뜻한 톤으로 표현되어, '얼굴이 파랗게 질린'이라는 지시사항이 반영되지 않았습니다.", "fix_en": "Change the character's facial skin tone to appear extremely pale and ashen, removing the warm hues to reflect severe fright." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the character's facial skin tone to appear extremely pale and ashen, removing the warm hues to reflect severe fright.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S14sh3" } }, "S15sh1::variants": { "author_fp": "25955330d9433289", "author": { "variants": [ { "approach_ko": "정면 아이레벨의 정적인 풀숏으로 인물의 고립감과 밤바다의 깊이를 균형 있게 담는다.", "prompt_en": "Photograph the moment as a restrained, static eye-level full shot, with her entire body and the worn bench clearly framed against the dark sea. Use a balanced frontal composition with ample breathing room around her, keeping the phone visible but visually subordinate. Shape the night illumination softly from one side, preserving natural facial detail beneath the bowed head while the background falls into deep, quiet tonal layers." }, { "approach_ko": "낮은 사선 원경에서 도로 가장자리를 전경에 두고, 넓은 밤바다의 여백 속에 인물을 작고 외롭게 배치한다.", "prompt_en": "Take a low, oblique long-shot view from near the open roadside edge, using the foreground to lead diagonally toward the bench while the sea occupies broad negative space beyond. Keep her full figure relatively small in the frame, with the bowed posture and phone readable through clean separation rather than close detail. Favor subdued back-side illumination that traces the figure and bench without turning either into a silhouette, leaving the surrounding night dense and isolating." } ] }, "reused": false }, "S15sh1": { "input_fingerprint": "aabdd4f40adb53e7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다를 배경으로 낡은 버스정류장 벤치에 고개를 숙인 채 멍하니 앉아 있는 수리영의 전신.\n\nLOCATION (lock): A bus-stop area along the coast with seating oriented toward the sea and an open roadside edge nearby. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still has her mobile phone and continues trying unsuccessfully to reach Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다를 배경으로 낡은 버스정류장 벤치에 고개를 숙인 채 멍하니 앉아 있는 수리영의 전신.\n\nLOCATION (lock): A bus-stop area along the coast with seating oriented toward the sea and an open roadside edge nearby. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still has her mobile phone and continues trying unsuccessfully to reach Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a restrained, static eye-level full shot, with her entire body and the worn bench clearly framed against the dark sea. Use a balanced frontal composition with ample breathing room around her, keeping the phone visible but visually subordinate. Shape the night illumination softly from one side, preserving natural facial detail beneath the bowed head while the background falls into deep, quiet tonal 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.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤바다를 배경으로 낡은 버스정류장 벤치에 고개를 숙인 채 멍하니 앉아 있는 수리영의 전신.\n\nLOCATION (lock): A bus-stop area along the coast with seating oriented toward the sea and an open roadside edge nearby. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still has her mobile phone and continues trying unsuccessfully to reach Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nTake a low, oblique long-shot view from near the open roadside edge, using the foreground to lead diagonally toward the bench while the sea occupies broad negative space beyond. Keep her full figure relatively small in the frame, with the bowed posture and phone readable through clean separation rather than close detail. Favor subdued back-side illumination that traces the figure and bench without turning either into a silhouette, leaving the surrounding night dense and isolating.\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 전신 샷, 밤바다 배경, 그리고 휴대폰을 든 상태를 정확히 구현했으며, 우측 구조물을 통해 버스정류장의 느낌을 잘 살렸습니다." }, { "label": "B", "score": 5, "verdict_ko": "지시된 조건들을 대략적으로 충족하나, 배경이 버스정류장보다는 단순한 해안가 벤치에 가깝게 연출되어 정확도가 다소 떨어집니다." } ], "refs": [ { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영이 누군가에게 연락을 시도하는 행동에 맞지 않게 휴대폰을 세로가 아닌 가로(가로 모드)로 쥐고 있습니다.", "fix_en": "Rotate the mobile phone in her hands so she is holding it vertically (portrait mode), which is appropriate for texting or calling." } ] }, "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 hands so she is holding it vertically (portrait mode), which is appropriate for texting or calling.\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", "share_plan": { "ref_plan": "background" } }, "S15sh4::variants": { "author_fp": "7042d20740a78b1e", "author": { "variants": [ { "approach_ko": "낮은 후방 근거리 시점과 넓은 원근감으로 달려 나가는 순간의 추진력을 강조한다.", "prompt_en": "Photograph from a low, close rear three-quarter position with a moderately wide lens feel, letting the roadway pull strongly into depth ahead of her. Use an off-center composition and a tight full-body crop that preserves the airborne foot, with slight natural motion blur at the extremities while her torso remains legible. Keep the night exposure low-key, using only the existing exterior illumination to trace her outline while retaining subtle clothing and body detail rather than rendering her as a featureless black shape." }, { "approach_ko": "멀리 떨어진 높은 후방 롱렌즈 시점으로 인물을 해안도로와 고정된 장소 안에 고립시킨다.", "prompt_en": "Use a distant, slightly elevated rear viewpoint with a compressed long-lens feel, holding her smaller within the established bus-stop and roadside geometry. Frame with strong negative space in the direction of movement and layered fixed elements between foreground, subject, and dark road, creating a restrained, observational tension. Expose for the night ambience so her back reads as a dark human figure with a delicate rim and faint tonal detail, sharply freezing the suspended mid-stride instant." } ] }, "reused": false }, "S15sh4": { "input_fingerprint": "604a5bdbc93ceda9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 어둠이 깔린 해안도로를 향해 한 발을 허공에 띄운 채 상체를 앞으로 기울인 달리기 mid-action 자세의 수리영 뒷모습 실루엣.\n\nLOCATION (lock): A bus-stop area along the coast with seating oriented toward the sea and an open roadside edge nearby. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong runs off with her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 어둠이 깔린 해안도로를 향해 한 발을 허공에 띄운 채 상체를 앞으로 기울인 달리기 mid-action 자세의 수리영 뒷모습 실루엣.\n\nLOCATION (lock): A bus-stop area along the coast with seating oriented toward the sea and an open roadside edge nearby. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong runs off with her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a low, close rear three-quarter position with a moderately wide lens feel, letting the roadway pull strongly into depth ahead of her. Use an off-center composition and a tight full-body crop that preserves the airborne foot, with slight natural motion blur at the extremities while her torso remains legible. Keep the night exposure low-key, using only the existing exterior illumination to trace her outline while retaining subtle clothing and body detail rather than rendering her as a featureless black 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.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, exterior illumination unspecified.\n\nSHOT TEXT (authoritative, Korean): 어둠이 깔린 해안도로를 향해 한 발을 허공에 띄운 채 상체를 앞으로 기울인 달리기 mid-action 자세의 수리영 뒷모습 실루엣.\n\nLOCATION (lock): A bus-stop area along the coast with seating oriented toward the sea and an open roadside edge nearby. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong runs off with her mobile phone 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a distant, slightly elevated rear viewpoint with a compressed long-lens feel, holding her smaller within the established bus-stop and roadside geometry. Frame with strong negative space in the direction of movement and layered fixed elements between foreground, subject, and dark road, creating a restrained, observational tension. Expose for the night ambience so her back reads as a dark human figure with a delicate rim and faint tonal detail, sharply freezing the suspended mid-stride instant.\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": "상체를 앞으로 기울인 역동적인 달리기 자세와 한 발을 띄운 뒷모습을 지시문대로 훌륭하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "상체가 거의 기울어지지 않아 요구된 달리기 자세의 역동성이 부족하며 정류장 구조도 레퍼런스와 차이가 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S15sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "캐릭터의 머리카락이 묶여 있습니다. 레퍼런스 이미지와 이전 샷에서는 머리를 풀고 있었습니다.", "fix_en": "Change the character's hair to be loose and flowing down her back instead of tied in a ponytail." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the character's hair to be loose and flowing down her back instead of tied in a ponytail.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S15sh1" } }, "S16sh1::variants": { "author_fp": "97903ec0d30d8d3a", "author": { "variants": [ { "approach_ko": "사무실 안쪽의 눈높이 정면 미디엄 클로즈업으로, 문을 밀고 들어오는 혜수의 긴박한 표정과 움직임을 공간 깊이 속에 포착한다.", "prompt_en": "Photograph from inside the office at eye level in a frontal medium close-up, catching Hye-su’s upper body as she crosses the doorway. Use a gently wide perspective so the door edge frames her while the desks and duty workstations recede behind her in layered depth; keep her furrowed face and properly held phone crisp, with restrained night-office illumination giving the entrance immediate urgency." }, { "approach_ko": "문 가까이의 비스듬한 측면 타이트 숏으로, 전화에 집중한 얼굴과 문을 미는 동작을 압축해 심리적 긴장을 강조한다.", "prompt_en": "Place the camera close beside the doorway for a tight oblique profile view, compressing Hye-su’s phone, furrowed face, and door-pushing gesture into one tense plane. Use a longer-lens feel and shallow depth of field, letting the police-office workstations fall softly out of focus while a narrow sideward wash of office light models her mature features and emphasizes her distracted forward step." } ] }, "reused": false }, "S16sh1": { "input_fingerprint": "6b204170f6104b87", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, office lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핸드폰을 귀에 댄 채 미간을 찌푸린 얼굴로 형사과 사무실 문을 밀며 한 발짝 내디딘 자세의 혜수 상체.\n\nLOCATION (lock): A police station office arranged with desks, computers, telephones, and duty workstations, largely empty except for the overnight officer. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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-su holds her mobile phone to her ear while calling Minsuk, who 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, office lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핸드폰을 귀에 댄 채 미간을 찌푸린 얼굴로 형사과 사무실 문을 밀며 한 발짝 내디딘 자세의 혜수 상체.\n\nLOCATION (lock): A police station office arranged with desks, computers, telephones, and duty workstations, largely empty except for the overnight officer. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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-su holds her mobile phone to her ear while calling Minsuk, who 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from inside the office at eye level in a frontal medium close-up, catching Hye-su’s upper body as she crosses the doorway. Use a gently wide perspective so the door edge frames her while the desks and duty workstations recede behind her in layered depth; keep her furrowed face and properly held phone crisp, with restrained night-office illumination giving the entrance immediate urgency.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, office lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핸드폰을 귀에 댄 채 미간을 찌푸린 얼굴로 형사과 사무실 문을 밀며 한 발짝 내디딘 자세의 혜수 상체.\n\nLOCATION (lock): A police station office arranged with desks, computers, telephones, and duty workstations, largely empty except for the overnight officer. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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-su holds her mobile phone to her ear while calling Minsuk, who 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera close beside the doorway for a tight oblique profile view, compressing Hye-su’s phone, furrowed face, and door-pushing gesture into one tense plane. Use a longer-lens feel and shallow depth of field, letting the police-office workstations fall softly out of focus while a narrow sideward wash of office light models her mature features and emphasizes her distracted forward step.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, office lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핸드폰을 귀에 댄 채 미간을 찌푸린 얼굴로 형사과 사무실 문을 밀며 한 발짝 내디딘 자세의 혜수 상체.\n\nLOCATION (lock): A police station office arranged with desks, computers, telephones, and duty workstations, largely empty except for the overnight officer. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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-su holds her mobile phone to her ear while calling Minsuk, who 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from inside the office at eye level in a frontal medium close-up, catching Hye-su’s upper body as she crosses the doorway. Use a gently wide perspective so the door edge frames her while the desks and duty workstations recede behind her in layered depth; keep her furrowed face and properly held phone crisp, with restrained night-office illumination giving the entrance immediate urgency.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, office lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 핸드폰을 귀에 댄 채 미간을 찌푸린 얼굴로 형사과 사무실 문을 밀며 한 발짝 내디딘 자세의 혜수 상체.\n\nLOCATION (lock): A police station office arranged with desks, computers, telephones, and duty workstations, largely empty except for the overnight officer. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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-su holds her mobile phone to her ear while calling Minsuk, who 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera close beside the doorway for a tight oblique profile view, compressing Hye-su’s phone, furrowed face, and door-pushing gesture into one tense plane. Use a longer-lens feel and shallow depth of field, letting the police-office workstations fall softly out of focus while a narrow sideward wash of office light models her mature features and emphasizes her distracted forward step.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S16sh1.png" }, { "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S16sh1.png" }, { "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B03.png" }, { "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B03.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "배경에 지시문에 없는 인물이 명확히 등장하여 규정을 위반했습니다." }, { "label": "B", "score": 6, "verdict_ko": "요구된 인물과 동작은 무난하게 구현되었으나 출입문의 형태가 장소와 다소 이질적입니다." }, { "label": "C", "score": 4, "verdict_ko": "모니터 뒤에 지시문에 없는 인물이 추가되어 하드 위반에 해당합니다." }, { "label": "D", "score": 7, "verdict_ko": "지정된 장소에서 문을 밀며 들어오는 인물의 동작과 표정을 가장 정확하게 묘사했습니다." } ] }, "forward_normalized": { "winner": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "배경에 지시문에 없는 인물이 명확히 등장하여 규정을 위반했습니다." }, { "label": "B", "score": 6, "verdict_ko": "요구된 인물과 동작은 무난하게 구현되었으나 출입문의 형태가 장소와 다소 이질적입니다." }, { "label": "C", "score": 4, "verdict_ko": "모니터 뒤에 지시문에 없는 인물이 추가되어 하드 위반에 해당합니다." }, { "label": "D", "score": 7, "verdict_ko": "지정된 장소에서 문을 밀며 들어오는 인물의 동작과 표정을 가장 정확하게 묘사했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "C", "B", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "샷 텍스트의 상체 프레이밍과 동작(문 밀기, 통화)을 명확히 구현했으며, 지시되지 않은 인물 추가 없이 기준을 가장 잘 충족함." }, { "label": "B", "score": 4, "verdict_ko": "모니터 뒤에 샷 텍스트에 없는 인물이 등장하고 화면에 텍스트가 포함되어 하드 위반에 해당함." }, { "label": "C", "score": 5, "verdict_ko": "프레이밍과 배경은 양호하나, 핸드폰의 뒷면(카메라 렌즈)을 귀에 대고 있어 소품 방향 규칙을 위반함." }, { "label": "D", "score": 3, "verdict_ko": "배경에 샷 텍스트에 명시되지 않은 인물이 뚜렷하게 서 있어 인물 추가 금지 하드 규정을 크게 위반함." } ] }, "reverse_normalized": { "winner": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "샷 텍스트의 상체 프레이밍과 동작(문 밀기, 통화)을 명확히 구현했으며, 지시되지 않은 인물 추가 없이 기준을 가장 잘 충족함." }, { "label": "C", "score": 4, "verdict_ko": "모니터 뒤에 샷 텍스트에 없는 인물이 등장하고 화면에 텍스트가 포함되어 하드 위반에 해당함." }, { "label": "B", "score": 5, "verdict_ko": "프레이밍과 배경은 양호하나, 핸드폰의 뒷면(카메라 렌즈)을 귀에 대고 있어 소품 방향 규칙을 위반함." }, { "label": "A", "score": 3, "verdict_ko": "배경에 샷 텍스트에 명시되지 않은 인물이 뚜렷하게 서 있어 인물 추가 금지 하드 규정을 크게 위반함." } ] }, "combined": { "totals": { "A": 6, "B": 11, "C": 8, "D": 14 }, "ranking": [ "D", "B", "C", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 11, "C": 8, "D": 14 }, "selected": "D", "ranking": [ "D", "B", "C", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "배경에 지시문에 없는 인물이 명확히 등장하여 규정을 위반했습니다." }, { "label": "B", "score": 6, "verdict_ko": "요구된 인물과 동작은 무난하게 구현되었으나 출입문의 형태가 장소와 다소 이질적입니다." }, { "label": "C", "score": 4, "verdict_ko": "모니터 뒤에 지시문에 없는 인물이 추가되어 하드 위반에 해당합니다." }, { "label": "D", "score": 7, "verdict_ko": "지정된 장소에서 문을 밀며 들어오는 인물의 동작과 표정을 가장 정확하게 묘사했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B03.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물의 오른쪽에 두 개의 팔(전화기를 든 팔과 문을 미는 팔)이 겹쳐서 생성되어 해부학적으로 심각한 오류가 있습니다.", "fix_en": "Remove the extra right arm and reassign the actions so the character has exactly one right arm and one left arm." }, { "issue_ko": "문을 미는 손의 엄지손가락이 아래쪽을 향하고 있어 사람의 손 구조와 맞지 않습니다.", "fix_en": "Redraw the hand pushing the door so the thumb is oriented correctly, pointing upwards." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the extra right arm and reassign the actions so the character has exactly one right arm and one left arm.\n- Redraw the hand pushing the door so the thumb is oriented correctly, pointing upwards.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L14B01", "B": "L14B02", "C": "L14B03" }, "assigned": "L14B03", "choice": "Candidate C", "confident": true, "reason_ko": "\"형사과 사무실 문을 밀며\"라는 지문으로 보아 사무실 출입구 쪽을 향한 동일한 공간(뷰)을 보여주며, 서브공간의 구분이 없으므로 현재 할당된 후보를 유지합니다.", "kept": "L14B03" } }, "S16sh4::variants": { "author_fp": "4fed015ae2a79e61", "author": { "variants": [ { "approach_ko": "김형사의 낮은 시선을 대신하는 정면 로우앵글 클로즈업으로 혜수의 단호한 압박감을 강조한다.", "prompt_en": "Frame Hye-su in a tight frontal close-up from slightly below her eye line, as if the camera occupies Detective Kim’s lower position. Keep her face dominant and nearly centered, with her firmly pressed lips and sharp downward gaze in precise focus; let the existing office environment recede into a soft, unobtrusive background. Use the established office light to model her mature features with restrained contrast and a controlled, confrontational stillness." }, { "approach_ko": "비스듬한 측면의 밀도 높은 얼굴 클로즈업과 얕은 심도로 혜수의 굳은 입술과 내리꽂는 시선을 포착한다.", "prompt_en": "Photograph Hye-su in a close three-quarter profile, placing the camera just off the downward path of her gaze rather than directly beneath it. Crop tightly around her face so the set of her jaw, compressed lips, and narrowed focus carry the frame, while a narrow suggestion of the established police office remains softly layered behind her. Hold the visible eye and lip line crisp, allowing the farther side of her face and background to fall away gently for a colder, more observational tension." } ] }, "reused": false }, "S16sh4": { "input_fingerprint": "ebf8990f2f40cffb", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, office lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 입술을 굳게 다문 채 단호하고 매서운 눈빛으로 김형사를 굽어보는 혜수의 얼굴 클로즈업.\n\nLOCATION (lock): A police station office arranged with desks, computers, telephones, and duty workstations, largely empty except for the overnight officer. 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 office lighting, institutional textures, and the investigator's continuing clothing and position near her colleague from the reference. Exclude the doorway-entry action and phone from the close frame, concentrating on her stern downward gaze.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-su continues holding the mobile phone containing Minsuk’s saved number.\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, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, office lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 입술을 굳게 다문 채 단호하고 매서운 눈빛으로 김형사를 굽어보는 혜수의 얼굴 클로즈업.\n\nLOCATION (lock): A police station office arranged with desks, computers, telephones, and duty workstations, largely empty except for the overnight officer. 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 office lighting, institutional textures, and the investigator's continuing clothing and position near her colleague from the reference. Exclude the doorway-entry action and phone from the close frame, concentrating on her stern downward gaze.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-su continues holding the mobile phone containing Minsuk’s saved number.\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, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame Hye-su in a tight frontal close-up from slightly below her eye line, as if the camera occupies Detective Kim’s lower position. Keep her face dominant and nearly centered, with her firmly pressed lips and sharp downward gaze in precise focus; let the existing office environment recede into a soft, unobtrusive background. Use the established office light to model her mature features with restrained contrast and a controlled, confrontational 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night, office lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 입술을 굳게 다문 채 단호하고 매서운 눈빛으로 김형사를 굽어보는 혜수의 얼굴 클로즈업.\n\nLOCATION (lock): A police station office arranged with desks, computers, telephones, and duty workstations, largely empty except for the overnight officer. 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 office lighting, institutional textures, and the investigator's continuing clothing and position near her colleague from the reference. Exclude the doorway-entry action and phone from the close frame, concentrating on her stern downward gaze.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-su continues holding the mobile phone containing Minsuk’s saved number.\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, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph Hye-su in a close three-quarter profile, placing the camera just off the downward path of her gaze rather than directly beneath it. Crop tightly around her face so the set of her jaw, compressed lips, and narrowed focus carry the frame, while a narrow suggestion of the established police office remains softly layered behind her. Hold the visible eye and lip line crisp, allowing the farther side of her face and background to fall away gently for a colder, more observational 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": 4, "B": 7 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "카메라를 정면으로 응시하고 있어 '굽어보는' 시선 지시를 위반했으며, 클로즈업에서 제외해야 할 휴대폰이 프레임에 포함되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "지시된 '굽어보는' 시선과 굳게 다문 입술의 표정을 훌륭하게 구현했으나, 프레임에서 제외할 것을 요구한 휴대폰이 여전히 노출되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S16sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "클로즈업 프레임에서 전화기를 제외하라는 지시를 어겼으며, 이전 숏에서 귀에 대고 있던 포즈와 달리 전화기를 아래로 내린 상태로 자세가 무단으로 변경되었습니다.", "fix_en": "Remove the hand and mobile phone from the bottom left corner, replacing them with the character's jacket and the office background." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the hand and mobile phone from the bottom left corner, replacing them with the character's jacket and the office background.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S16sh1" } }, "S17sh2::variants": { "author_fp": "e9458b1aa462baa2", "author": { "variants": [ { "approach_ko": "고양이와 같은 높이의 절제된 측면 관찰숏으로, 응시선 끝의 창문을 압축해 긴장감을 만든다.", "prompt_en": "Photograph from the cat’s height in a restrained side-on medium-long view, preserving the complete profile and all four paws with clean separation against the surroundings. Use a gently compressed perspective so the illuminated rooftop-room window sits clearly along the cat’s unwavering sightline; keep the cat sharply resolved while allowing the room and rooftop depth to fall just slightly softer. Let the dawn exterior remain subdued against the contained interior glow, creating quiet, watchful tension." }, { "approach_ko": "낮고 넓은 환경 숏으로 낡은 담장을 강한 유도선으로 삼아 고양이와 옥탑방의 공간적 관계를 강조한다.", "prompt_en": "Use a wider, low-positioned environmental composition, letting the old boundary wall run prominently through the frame as a strong directional line toward the rooftop dwelling. Hold both the cat’s full lateral stance and the illuminated window in clear spatial relation with deeper focus, emphasizing the exposed rooftop geometry and the still distance between watcher and destination. Balance the cool dawn ambience with the window’s warmer interior light without turning either into an exaggerated effect." } ] }, "reused": false }, "S17sh2": { "input_fingerprint": "2862782be35a4f4b", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn, rooftop apartment light visible.\n\nSHOT TEXT (authoritative, Korean): 낡은 담장 위에 네 발을 딛고 서서 불 켜진 옥탑방 창문을 빤히 주시하는 검은 고양이의 측면 전신.\n\nLOCATION (lock): The rooftop of an old residential building with a small rooftop dwelling and a boundary wall crossed by a cat. The dwelling's interior light is visible from outside. 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 remains lit, while its interior retains the unnaturally clean state left after the corpse and evidence vanished.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn, rooftop apartment light visible.\n\nSHOT TEXT (authoritative, Korean): 낡은 담장 위에 네 발을 딛고 서서 불 켜진 옥탑방 창문을 빤히 주시하는 검은 고양이의 측면 전신.\n\nLOCATION (lock): The rooftop of an old residential building with a small rooftop dwelling and a boundary wall crossed by a cat. The dwelling's interior light is visible from outside. 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 remains lit, while its interior retains the unnaturally clean state left after the corpse and evidence vanished.\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 the cat’s height in a restrained side-on medium-long view, preserving the complete profile and all four paws with clean separation against the surroundings. Use a gently compressed perspective so the illuminated rooftop-room window sits clearly along the cat’s unwavering sightline; keep the cat sharply resolved while allowing the room and rooftop depth to fall just slightly softer. Let the dawn exterior remain subdued against the contained interior glow, creating quiet, watchful 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn, rooftop apartment light visible.\n\nSHOT TEXT (authoritative, Korean): 낡은 담장 위에 네 발을 딛고 서서 불 켜진 옥탑방 창문을 빤히 주시하는 검은 고양이의 측면 전신.\n\nLOCATION (lock): The rooftop of an old residential building with a small rooftop dwelling and a boundary wall crossed by a cat. The dwelling's interior light is visible from outside. 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 remains lit, while its interior retains the unnaturally clean state left after the corpse and evidence vanished.\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, low-positioned environmental composition, letting the old boundary wall run prominently through the frame as a strong directional line toward the rooftop dwelling. Hold both the cat’s full lateral stance and the illuminated window in clear spatial relation with deeper focus, emphasizing the exposed rooftop geometry and the still distance between watcher and destination. Balance the cool dawn ambience with the window’s warmer interior light without turning either into an exaggerated effect.\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": "고양이가 서 있는 모습은 구현되었으나, 참조 이미지와 달리 옥탑방 문이 통유리로 변형되어 빛이 새어나오는 구조적 오류가 있습니다." }, { "label": "B", "score": 7, "verdict_ko": "담장 위에 네 발로 서서 불 켜진 창문을 바라보는 고양이의 측면 전신을 정확히 묘사했으며, 옥탑방의 구조와 조명 상태도 참조 이미지와 일치합니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L12B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/background_chain/seed_bg_rooftop_villa_complex_sel.png" } ], "critique": { "issues": [ { "issue_ko": "옥탑방 건물의 외관(창살 있는 큰 창문과 철문, 외벽 재질)이 STRUCTURE LOOK 사진이 아닌 LOCATION 사진을 그대로 따르고 있습니다.", "fix_en": "Change the rooftop dwelling's exterior to match the STRUCTURE LOOK photo, replacing the large barred window and metal door with the green door, small white window, and concrete walls." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the rooftop dwelling's exterior to match the STRUCTURE LOOK photo, replacing the large barred window and metal door with the green door, small white window, and concrete walls.\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)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only", "plate_select": { "candidates": { "A": "L12B01", "B": "L12B02", "C": "L12B03", "D": "L12B04" }, "assigned": "L12B01", "choice": "Candidate A", "confident": true, "reason_ko": "\"낡은 담장 위에... 불 켜진 옥탑방 창문을 빤히 주시하는 검은 고양이\"라는 묘사에 정확히 부합하게 담장 위의 고양이와 불 켜진 옥탑방 창문이 한 프레임에 담겨 있는 공간이므로 Candidate A가 가장 적합합니다.", "kept": "L12B01" } }, "S18sh2::variants": { "author_fp": "2338c31f029e7a4f", "author": { "variants": [ { "approach_ko": "정면 눈높이의 타이트한 상반신 클로즈업으로 수리영의 돌연 굳은 긴장과 크게 뜬 눈에 집중한다.", "prompt_en": "Frame a tight, straight-on upper-body close-up at eye level, with her face held near the center in restrained symmetry. Use shallow depth of field to isolate the sharply focused eyes and rigid posture, while soft early-morning ambient light gives the stillness an immediate, intimate tension." }, { "approach_ko": "조금 낮은 정면 시점의 넓은 미디엄 클로즈업으로 꼿꼿한 자세와 주변의 정적을 함께 강조한다.", "prompt_en": "Photograph her frontally in a wider medium close-up from slightly below chest level, preserving more space around her upright torso so the rigid vertical posture dominates the composition. Keep the surrounding room legible with deeper focus and let directional early-morning light create controlled planes across her face, making her frozen alertness feel exposed within the unnatural quiet." } ] }, "reused": false }, "S18sh2": { "input_fingerprint": "700a69769ae7dda2", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 두 눈을 번쩍 뜨고 상체를 꼿꼿이 세운 채 긴장한 표정으로 멈춘 수리영의 정면 상체.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): The rooftop room remains unnaturally clean and empty of Minsuk, and Su-ri-yeong still has her mobile phone nearby after waiting through the night.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 두 눈을 번쩍 뜨고 상체를 꼿꼿이 세운 채 긴장한 표정으로 멈춘 수리영의 정면 상체.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): The rooftop room remains unnaturally clean and empty of Minsuk, and Su-ri-yeong still has her mobile phone nearby after waiting through the night.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight, straight-on upper-body close-up at eye level, with her face held near the center in restrained symmetry. Use shallow depth of field to isolate the sharply focused eyes and rigid posture, while soft early-morning ambient light gives the stillness an immediate, intimate 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 두 눈을 번쩍 뜨고 상체를 꼿꼿이 세운 채 긴장한 표정으로 멈춘 수리영의 정면 상체.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): The rooftop room remains unnaturally clean and empty of Minsuk, and Su-ri-yeong still has her mobile phone nearby after waiting through the night.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her frontally in a wider medium close-up from slightly below chest level, preserving more space around her upright torso so the rigid vertical posture dominates the composition. Keep the surrounding room legible with deeper focus and let directional early-morning light create controlled planes across her face, making her frozen alertness feel exposed within the unnatural quiet.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 두 눈을 번쩍 뜨고 상체를 꼿꼿이 세운 채 긴장한 표정으로 멈춘 수리영의 정면 상체.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): The rooftop room remains unnaturally clean and empty of Minsuk, and Su-ri-yeong still has her mobile phone nearby after waiting through the night.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight, straight-on upper-body close-up at eye level, with her face held near the center in restrained symmetry. Use shallow depth of field to isolate the sharply focused eyes and rigid posture, while soft early-morning ambient light gives the stillness an immediate, intimate 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 두 눈을 번쩍 뜨고 상체를 꼿꼿이 세운 채 긴장한 표정으로 멈춘 수리영의 정면 상체.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): The rooftop room remains unnaturally clean and empty of Minsuk, and Su-ri-yeong still has her mobile phone nearby after waiting through the night.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her frontally in a wider medium close-up from slightly below chest level, preserving more space around her upright torso so the rigid vertical posture dominates the composition. Keep the surrounding room legible with deeper focus and let directional early-morning light create controlled planes across her face, making her frozen alertness feel exposed within the unnatural quiet.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S18sh2.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S18sh2.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B03.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": "A", "score": 5, "verdict_ko": "지정된 장소와 구도를 잘 구현했으나, 캐릭터 레퍼런스의 어두운 재킷을 누락했습니다." }, { "label": "B", "score": 4, "verdict_ko": "장소와 표정은 적절하나, 레퍼런스와 전혀 다른 디자인의 긴팔 의상을 입고 있습니다." }, { "label": "C", "score": 3, "verdict_ko": "캐릭터 의상은 일치하지만, 지정된 장소 레퍼런스의 배경 및 공간적 특징을 반영하지 못했습니다." }, { "label": "D", "score": 7, "verdict_ko": "캐릭터의 의상(재킷 포함), 지정된 위치, 정면 상체 구도 및 긴장한 표정을 가장 정확하게 재현했습니다." } ] }, "forward_normalized": { "winner": "D", "ranking": [ "D", "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "지정된 장소와 구도를 잘 구현했으나, 캐릭터 레퍼런스의 어두운 재킷을 누락했습니다." }, { "label": "B", "score": 4, "verdict_ko": "장소와 표정은 적절하나, 레퍼런스와 전혀 다른 디자인의 긴팔 의상을 입고 있습니다." }, { "label": "C", "score": 3, "verdict_ko": "캐릭터 의상은 일치하지만, 지정된 장소 레퍼런스의 배경 및 공간적 특징을 반영하지 못했습니다." }, { "label": "D", "score": 7, "verdict_ko": "캐릭터의 의상(재킷 포함), 지정된 위치, 정면 상체 구도 및 긴장한 표정을 가장 정확하게 재현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "D", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "정면 상체 프레이밍, 두 눈을 뜬 긴장한 표정, 지정된 로케이션과 의상(재킷)을 모두 충실하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "인물의 구도와 의상은 적절하나, 지정된 장소의 고유한 공간적 특징(배경)이 화면에 나타나지 않았습니다." }, { "label": "C", "score": 4, "verdict_ko": "장소와 프레이밍은 지침을 잘 따랐으나, 레퍼런스와 전혀 다른 의상을 입고 있습니다." }, { "label": "D", "score": 5, "verdict_ko": "지정된 장소와 상체 구도를 정확히 묘사했으나, 레퍼런스의 재킷이 누락되었습니다." } ] }, "reverse_normalized": { "winner": "D", "ranking": [ "D", "A", "B", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "정면 상체 프레이밍, 두 눈을 뜬 긴장한 표정, 지정된 로케이션과 의상(재킷)을 모두 충실하게 구현했습니다." }, { "label": "C", "score": 3, "verdict_ko": "인물의 구도와 의상은 적절하나, 지정된 장소의 고유한 공간적 특징(배경)이 화면에 나타나지 않았습니다." }, { "label": "B", "score": 4, "verdict_ko": "장소와 프레이밍은 지침을 잘 따랐으나, 레퍼런스와 전혀 다른 의상을 입고 있습니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 장소와 상체 구도를 정확히 묘사했으나, 레퍼런스의 재킷이 누락되었습니다." } ] }, "combined": { "totals": { "A": 10, "B": 8, "C": 6, "D": 14 }, "ranking": [ "D", "A", "B", "C" ], "agreement": true, "policy": 1 } }, "totals": { "A": 10, "B": 8, "C": 6, "D": 14 }, "selected": "D", "ranking": [ "D", "A", "B", "C" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "지정된 장소와 구도를 잘 구현했으나, 캐릭터 레퍼런스의 어두운 재킷을 누락했습니다." }, { "label": "B", "score": 4, "verdict_ko": "장소와 표정은 적절하나, 레퍼런스와 전혀 다른 디자인의 긴팔 의상을 입고 있습니다." }, { "label": "C", "score": 3, "verdict_ko": "캐릭터 의상은 일치하지만, 지정된 장소 레퍼런스의 배경 및 공간적 특징을 반영하지 못했습니다." }, { "label": "D", "score": 7, "verdict_ko": "캐릭터의 의상(재킷 포함), 지정된 위치, 정면 상체 구도 및 긴장한 표정을 가장 정확하게 재현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B03.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 '두 눈을 번쩍 뜨고'라고 명시되어 있으나, 캐릭터의 눈이 이완되어 평범하게 떠 있습니다.", "fix_en": "Open the character's eyes wide to clearly show a tense, startled expression." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Open the character's eyes wide to clearly show a tense, startled expression.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B03", "choice": "Candidate C", "confident": true, "reason_ko": "\"상체를 꼿꼿이 세운 채\"라는 묘사는 잠자리에서 갑자기 일어나는 상황을 암시하므로, 침대가 위치한 현재 할당된 침실 뷰가 가장 적합합니다.", "kept": "L05B03" } }, "S18sh8::variants": { "author_fp": "73e255ab75f141a9", "author": { "variants": [ { "approach_ko": "수리영의 어깨를 크게 전경에 둔 밀착 오버숄더로, 거울 속 붉은 원형 표식에만 시선을 고정한다.", "prompt_en": "Use a tight, eye-level over-the-shoulder composition, with Su-ri-yeong’s shoulder occupying a substantial soft-edged foreground and the mirror reflection dominating the frame. Hold crisp focus on the vivid red circular mark against the reflected bathroom tiles while letting the near shoulder fall gently out of focus. Keep the camera static and the illumination cool, soft, and restrained, creating an intimate moment of recognition." }, { "approach_ko": "거울과 타일의 직선 구조를 정면적이고 깊은 초점으로 배열해 붉은 원을 불길한 중심축으로 만든다.", "prompt_en": "Photograph the moment as a wider, formally locked composition from just behind Su-ri-yeong, using the mirror frame and tiled grid as strong rectilinear geometry. Keep her shoulder at the edge of the foreground while preserving clear depth through the mirror to the reflected wall, placing the vivid red circle as the composition’s stark visual anchor. Use even, low-contrast early-morning illumination and a still, observational camera to make the mark feel disturbingly undeniable." } ] }, "reused": false }, "S18sh8": { "input_fingerprint": "c13b56051c66dd93", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 거울에 반사된 욕실 타일 벽면에 피처럼 붉은색 원형 표식이 선명하게 떠오른 정지 구도.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): Su-ri-yeong now carries her doll-adorned backpack and still has her mobile phone. A vivid red circle is temporarily visible on the reflected bathroom 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 거울에 반사된 욕실 타일 벽면에 피처럼 붉은색 원형 표식이 선명하게 떠오른 정지 구도.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): Su-ri-yeong now carries her doll-adorned backpack and still has her mobile phone. A vivid red circle is temporarily visible on the reflected bathroom 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a tight, eye-level over-the-shoulder composition, with Su-ri-yeong’s shoulder occupying a substantial soft-edged foreground and the mirror reflection dominating the frame. Hold crisp focus on the vivid red circular mark against the reflected bathroom tiles while letting the near shoulder fall gently out of focus. Keep the camera static and the illumination cool, soft, and restrained, creating an intimate moment of recognition.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 거울에 반사된 욕실 타일 벽면에 피처럼 붉은색 원형 표식이 선명하게 떠오른 정지 구도.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): Su-ri-yeong now carries her doll-adorned backpack and still has her mobile phone. A vivid red circle is temporarily visible on the reflected bathroom 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a wider, formally locked composition from just behind Su-ri-yeong, using the mirror frame and tiled grid as strong rectilinear geometry. Keep her shoulder at the edge of the foreground while preserving clear depth through the mirror to the reflected wall, placing the vivid red circle as the composition’s stark visual anchor. Use even, low-contrast early-morning illumination and a still, observational camera to make the mark feel disturbingly undeniable.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 거울에 반사된 욕실 타일 벽면에 피처럼 붉은색 원형 표식이 선명하게 떠오른 정지 구도.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): Su-ri-yeong now carries her doll-adorned backpack and still has her mobile phone. A vivid red circle is temporarily visible on the reflected bathroom 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a tight, eye-level over-the-shoulder composition, with Su-ri-yeong’s shoulder occupying a substantial soft-edged foreground and the mirror reflection dominating the frame. Hold crisp focus on the vivid red circular mark against the reflected bathroom tiles while letting the near shoulder fall gently out of focus. Keep the camera static and the illumination cool, soft, and restrained, creating an intimate moment of recognition.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 수리영의 어깨 너머로 거울에 반사된 욕실 타일 벽면에 피처럼 붉은색 원형 표식이 선명하게 떠오른 정지 구도.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): Su-ri-yeong now carries her doll-adorned backpack and still has her mobile phone. A vivid red circle is temporarily visible on the reflected bathroom 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: 수리영 (한국인 여성, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as a wider, formally locked composition from just behind Su-ri-yeong, using the mirror frame and tiled grid as strong rectilinear geometry. Keep her shoulder at the edge of the foreground while preserving clear depth through the mirror to the reflected wall, placing the vivid red circle as the composition’s stark visual anchor. Use even, low-contrast early-morning illumination and a still, observational camera to make the mark feel disturbingly undeniable.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S18sh8.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S18sh8.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B01.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": "어깨 너머 구도를 완벽히 구현했으며, 휴대폰과 인형 배낭 등 소지품 상태와 거울 속 붉은 표식, 배경 디테일이 가장 정확합니다." }, { "label": "B", "score": 4, "verdict_ko": "측면 수납장 거울에 얼굴이 정면으로 반사되는 물리적으로 불가능한 광학적 오류가 있어 감점되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "거실과 욕실의 구조가 비현실적으로 병합되어 공간의 물리적 형태가 심각하게 왜곡된 하드 위반입니다." }, { "label": "D", "score": 6, "verdict_ko": "공간과 구도는 준수하나, 화면 내에 휴대폰이 제외되었고 배낭의 장식 디테일이 A 후보에 비해 부족합니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "D", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "어깨 너머 구도를 완벽히 구현했으며, 휴대폰과 인형 배낭 등 소지품 상태와 거울 속 붉은 표식, 배경 디테일이 가장 정확합니다." }, { "label": "B", "score": 4, "verdict_ko": "측면 수납장 거울에 얼굴이 정면으로 반사되는 물리적으로 불가능한 광학적 오류가 있어 감점되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "거실과 욕실의 구조가 비현실적으로 병합되어 공간의 물리적 형태가 심각하게 왜곡된 하드 위반입니다." }, { "label": "D", "score": 6, "verdict_ko": "공간과 구도는 준수하나, 화면 내에 휴대폰이 제외되었고 배낭의 장식 디테일이 A 후보에 비해 부족합니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "A", "B", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "어깨 너머 구도, 소지품, 거울 속 붉은 표식과 공간 구조를 모두 정확히 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "구도는 적절하나 거울에 존재하지 않는 창문이 반사되고 인물 반사가 누락되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "레퍼런스 사진의 위치 표시용 마커(거실 벽)까지 그대로 화면에 유출되어 실격입니다." }, { "label": "C", "score": 3, "verdict_ko": "두 개의 거울에 물리적으로 불가능한 각도의 인물 얼굴이 반사되어 실격입니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "D", "C", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "어깨 너머 구도, 소지품, 거울 속 붉은 표식과 공간 구조를 모두 정확히 구현했습니다." }, { "label": "D", "score": 6, "verdict_ko": "구도는 적절하나 거울에 존재하지 않는 창문이 반사되고 인물 반사가 누락되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "레퍼런스 사진의 위치 표시용 마커(거실 벽)까지 그대로 화면에 유출되어 실격입니다." }, { "label": "B", "score": 3, "verdict_ko": "두 개의 거울에 물리적으로 불가능한 각도의 인물 얼굴이 반사되어 실격입니다." } ] }, "combined": { "totals": { "A": 14, "B": 7, "C": 7, "D": 12 }, "ranking": [ "A", "D", "B", "C" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 7, "C": 7, "D": 12 }, "selected": "A", "ranking": [ "A", "D", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "어깨 너머 구도를 완벽히 구현했으며, 휴대폰과 인형 배낭 등 소지품 상태와 거울 속 붉은 표식, 배경 디테일이 가장 정확합니다." }, { "label": "B", "score": 4, "verdict_ko": "측면 수납장 거울에 얼굴이 정면으로 반사되는 물리적으로 불가능한 광학적 오류가 있어 감점되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "거실과 욕실의 구조가 비현실적으로 병합되어 공간의 물리적 형태가 심각하게 왜곡된 하드 위반입니다." }, { "label": "D", "score": 6, "verdict_ko": "공간과 구도는 준수하나, 화면 내에 휴대폰이 제외되었고 배낭의 장식 디테일이 A 후보에 비해 부족합니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "휴대전화의 화면이 캐릭터의 시선 방향이 아닌 바깥쪽(욕실 방향)을 향하고 있으며, 폰을 쥔 오른손의 엄지와 손가락 위치가 뒤바뀌어 해부학적으로 틀렸습니다.", "fix_en": "Redraw the right hand with correct anatomy, and rotate the phone so its screen faces the character's eyes, showing its back to the camera." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw the right hand with correct anatomy, and rotate the phone so its screen faces the character's eyes, showing its back to the camera.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B01", "choice": "A", "confident": true, "reason_ko": "거울에 반사된 욕실 타일 벽면이 명시되어 있으며, 해당 조건을 만족하는 욕실 거울과 타일이 보이는 이미지는 A입니다.", "chosen": "L05B01" } }, "S18sh10::variants": { "author_fp": "183b62bfb4456b22", "author": { "variants": [ { "approach_ko": "액자 정면에 밀착한 극단적 클로즈업으로 웃는 얼굴의 붕괴를 냉정하고 대칭적으로 포착한다.", "prompt_en": "Photograph it as a locked, nearly straight-on extreme close-up, with the framed image filling almost the entire composition and the table reduced to a narrow grounding edge. Keep the frame geometry rigid and controlled around the grotesquely yielding face, using crisp focus across the photograph and soft, neutral early-morning illumination to make the impossible change feel like a practical effect captured in camera." }, { "approach_ko": "식탁 높이의 비스듬한 근접 시점과 얕은 심도로 액자의 물성과 흘러내리는 얼굴을 불안하게 강조한다.", "prompt_en": "Shoot from table height at a close oblique angle, letting the picture frame cut diagonally through the composition while its near edge and the melting facial image occupy distinct depth planes. Use very shallow focus centered on the distorted face, with restrained early-morning side light grazing the frame and printed surface so the deformation feels tactile, immediate, and physically present." } ] }, "reused": false }, "S18sh10": { "input_fingerprint": "44084ff7b51beceb", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 거실 식탁 위에 놓인 사진 액자 속 환하게 웃는 민숙의 얼굴 이미지가 찰흙처럼 기괴하게 일그러져 흘러내리는 찰나의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): Su-ri-yeong continues carrying the doll-adorned backpack and her phone. The framed photograph remains on the table as Minsuk’s pictured face briefly distorts and melts.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 거실 식탁 위에 놓인 사진 액자 속 환하게 웃는 민숙의 얼굴 이미지가 찰흙처럼 기괴하게 일그러져 흘러내리는 찰나의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): Su-ri-yeong continues carrying the doll-adorned backpack and her phone. The framed photograph remains on the table as Minsuk’s pictured face briefly distorts and melts.\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 locked, nearly straight-on extreme close-up, with the framed image filling almost the entire composition and the table reduced to a narrow grounding edge. Keep the frame geometry rigid and controlled around the grotesquely yielding face, using crisp focus across the photograph and soft, neutral early-morning illumination to make the impossible change feel like a practical effect captured in camera.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 거실 식탁 위에 놓인 사진 액자 속 환하게 웃는 민숙의 얼굴 이미지가 찰흙처럼 기괴하게 일그러져 흘러내리는 찰나의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment containing a main bedroom, a separate bedroom with a curtained window, a 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): Su-ri-yeong continues carrying the doll-adorned backpack and her phone. The framed photograph remains on the table as Minsuk’s pictured face briefly distorts and melts.\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):\nShoot from table height at a close oblique angle, letting the picture frame cut diagonally through the composition while its near edge and the melting facial image occupy distinct depth planes. Use very shallow focus centered on the distorted face, with restrained early-morning side light grazing the frame and printed surface so the deformation feels tactile, immediate, and physically present.\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": 9 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "클로즈업과 찰흙처럼 흘러내리는 효과는 좋으나, '민숙'이라는 이름에 어울리지 않게 남성의 얼굴이 등장했습니다." }, { "label": "B", "score": 9, "verdict_ko": "여성(민숙)의 얼굴이 기괴하게 일그러지며 찰흙처럼 흘러내리는 묘사를 클로즈업으로 가장 정확하고 생생하게 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B01.png" } ], "critique": { "issues": [] }, "fix_skipped": true, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B01", "choice": "Candidate B", "confident": true, "reason_ko": "원문에서 '거실 식탁 위에 놓인 사진 액자'를 명시하고 있으므로, 욕실이 중심인 현재 할당본(A) 대신 거실과 식탁이 온전히 담긴 서브공간(B)으로 변경해야 합니다.", "kept": "L05B01" } }, "S19sh4::variants": { "author_fp": "692e3cfa6a9108d1", "author": { "variants": [ { "approach_ko": "긴 렌즈의 측면 미디엄 클로즈업으로 수리영의 시선과 CCTV를 한 평면에 압축한다.", "prompt_en": "Photograph a restrained side-profile medium close-up from her eye level, using a long-lens feel to compress her face and the CCTV camera into the same shallow visual plane. Place her profile prominently to one side and the camera near the opposite upper edge, with her lifted gaze forming the compositional link; retain enough of her shoulder and lowered hand to register what she carries. Use soft early-morning ambient light across her features and selective focus that keeps her eye and the CCTV housing crisp while the facade falls gently away." }, { "approach_ko": "벽 가까이의 낮은 광각 시점에서 CCTV를 전경에 두고 아래의 수리영이 올려다보는 관계를 강조한다.", "prompt_en": "Shoot from a low position close to the exterior wall with a wide, spatial lens feel, letting the fixed CCTV camera dominate the near upper foreground while Su-ri-yeong appears below in clean side profile looking up at it. Build the frame on a strong diagonal between her face and the camera, preserving visible depth along the facade and including her carried belongings within the lower composition. Keep the early-morning light cool and even, with deeper focus so both her profile and the surveillance camera remain clearly legible." } ] }, "reused": false }, "S19sh4": { "input_fingerprint": "f6cc8fd53f58ae9e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 마트 외부 건물 벽에 달린 CCTV 카메라를 올려다보는 수리영의 옆얼굴.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with merchandise aisles and employee areas, plus an exterior bicycle parking area and a facade covered by security 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): Su-ri-yeong’s bicycle remains parked outside the mart, and she continues carrying her doll-adorned backpack and mobile phone. The exterior CCTV camera stays fixed in the identified position.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 마트 외부 건물 벽에 달린 CCTV 카메라를 올려다보는 수리영의 옆얼굴.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with merchandise aisles and employee areas, plus an exterior bicycle parking area and a facade covered by security 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): Su-ri-yeong’s bicycle remains parked outside the mart, and she continues carrying her doll-adorned backpack and mobile phone. The exterior CCTV camera stays fixed in the identified position.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph a restrained side-profile medium close-up from her eye level, using a long-lens feel to compress her face and the CCTV camera into the same shallow visual plane. Place her profile prominently to one side and the camera near the opposite upper edge, with her lifted gaze forming the compositional link; retain enough of her shoulder and lowered hand to register what she carries. Use soft early-morning ambient light across her features and selective focus that keeps her eye and the CCTV housing crisp while the facade falls gently away.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 마트 외부 건물 벽에 달린 CCTV 카메라를 올려다보는 수리영의 옆얼굴.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with merchandise aisles and employee areas, plus an exterior bicycle parking area and a facade covered by security 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): Su-ri-yeong’s bicycle remains parked outside the mart, and she continues carrying her doll-adorned backpack and mobile phone. The exterior CCTV camera stays fixed in the identified position.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from a low position close to the exterior wall with a wide, spatial lens feel, letting the fixed CCTV camera dominate the near upper foreground while Su-ri-yeong appears below in clean side profile looking up at it. Build the frame on a strong diagonal between her face and the camera, preserving visible depth along the facade and including her carried belongings within the lower composition. Keep the early-morning light cool and even, with deeper focus so both her profile and the surveillance camera remain clearly legible.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 마트 외부 건물 벽에 달린 CCTV 카메라를 올려다보는 수리영의 옆얼굴.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with merchandise aisles and employee areas, plus an exterior bicycle parking area and a facade covered by security 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): Su-ri-yeong’s bicycle remains parked outside the mart, and she continues carrying her doll-adorned backpack and mobile phone. The exterior CCTV camera stays fixed in the identified position.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph a restrained side-profile medium close-up from her eye level, using a long-lens feel to compress her face and the CCTV camera into the same shallow visual plane. Place her profile prominently to one side and the camera near the opposite upper edge, with her lifted gaze forming the compositional link; retain enough of her shoulder and lowered hand to register what she carries. Use soft early-morning ambient light across her features and selective focus that keeps her eye and the CCTV housing crisp while the facade falls gently away.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 마트 외부 건물 벽에 달린 CCTV 카메라를 올려다보는 수리영의 옆얼굴.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with merchandise aisles and employee areas, plus an exterior bicycle parking area and a facade covered by security 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): Su-ri-yeong’s bicycle remains parked outside the mart, and she continues carrying her doll-adorned backpack and mobile phone. The exterior CCTV camera stays fixed in the identified position.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from a low position close to the exterior wall with a wide, spatial lens feel, letting the fixed CCTV camera dominate the near upper foreground while Su-ri-yeong appears below in clean side profile looking up at it. Build the frame on a strong diagonal between her face and the camera, preserving visible depth along the facade and including her carried belongings within the lower composition. Keep the early-morning light cool and even, with deeper focus so both her profile and the surveillance camera remain clearly legible.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B03.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B03.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": "'옆얼굴'을 강조하는 알맞은 프레이밍을 사용했으며, CCTV의 위치와 인물의 시선이 레퍼런스와 가장 정확하게 일치합니다." }, { "label": "B", "score": 6, "verdict_ko": "요소들은 잘 갖춰졌으나 샷이 다소 넓게 잡혀 '옆얼굴'에 집중하는 샷 텍스트의 의도에 비해 프레이밍이 아쉽습니다." }, { "label": "C", "score": 4, "verdict_ko": "CCTV의 장착 위치가 레퍼런스 공간과 다르게 왜곡되었으며, 간판에 불필요한 문자가 추가되는 위반이 발생했습니다." }, { "label": "D", "score": 3, "verdict_ko": "CCTV 기기에 금지된 텍스트가 생성되었으며, 카메라의 크기가 과장되고 스마트폰을 쥔 손의 해부학적 오류가 큽니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'옆얼굴'을 강조하는 알맞은 프레이밍을 사용했으며, CCTV의 위치와 인물의 시선이 레퍼런스와 가장 정확하게 일치합니다." }, { "label": "B", "score": 6, "verdict_ko": "요소들은 잘 갖춰졌으나 샷이 다소 넓게 잡혀 '옆얼굴'에 집중하는 샷 텍스트의 의도에 비해 프레이밍이 아쉽습니다." }, { "label": "C", "score": 4, "verdict_ko": "CCTV의 장착 위치가 레퍼런스 공간과 다르게 왜곡되었으며, 간판에 불필요한 문자가 추가되는 위반이 발생했습니다." }, { "label": "D", "score": 3, "verdict_ko": "CCTV 기기에 금지된 텍스트가 생성되었으며, 카메라의 크기가 과장되고 스마트폰을 쥔 손의 해부학적 오류가 큽니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "A", "C", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "'옆얼굴'을 부각하는 적절한 프레이밍을 보여주며, 고정된 CCTV 위치와 소지품 등 모든 지침을 충실히 이행했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 디테일은 잘 유지했으나, 프레임이 불필요하게 넓어져 '옆얼굴'을 강조하는 샷의 의도가 다소 희석되었습니다." }, { "label": "C", "score": 5, "verdict_ko": "A와 마찬가지로 구도가 너무 넓어 피사체의 얼굴보다 배경이 더 많이 보이게 되어 감점 요인이 되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "얼굴 중심의 구도는 좋으나, 지정된 위치에 고정되어야 할 CCTV가 엉뚱한 벽면으로 이동되어 공간 지침을 크게 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "D", "B", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'옆얼굴'을 부각하는 적절한 프레이밍을 보여주며, 고정된 CCTV 위치와 소지품 등 모든 지침을 충실히 이행했습니다." }, { "label": "D", "score": 5, "verdict_ko": "지정된 디테일은 잘 유지했으나, 프레임이 불필요하게 넓어져 '옆얼굴'을 강조하는 샷의 의도가 다소 희석되었습니다." }, { "label": "B", "score": 5, "verdict_ko": "A와 마찬가지로 구도가 너무 넓어 피사체의 얼굴보다 배경이 더 많이 보이게 되어 감점 요인이 되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "얼굴 중심의 구도는 좋으나, 지정된 위치에 고정되어야 할 CCTV가 엉뚱한 벽면으로 이동되어 공간 지침을 크게 위반했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 11, "C": 7, "D": 8 }, "ranking": [ "A", "B", "D", "C" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 11, "C": 7, "D": 8 }, "selected": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'옆얼굴'을 강조하는 알맞은 프레이밍을 사용했으며, CCTV의 위치와 인물의 시선이 레퍼런스와 가장 정확하게 일치합니다." }, { "label": "B", "score": 6, "verdict_ko": "요소들은 잘 갖춰졌으나 샷이 다소 넓게 잡혀 '옆얼굴'에 집중하는 샷 텍스트의 의도에 비해 프레이밍이 아쉽습니다." }, { "label": "C", "score": 4, "verdict_ko": "CCTV의 장착 위치가 레퍼런스 공간과 다르게 왜곡되었으며, 간판에 불필요한 문자가 추가되는 위반이 발생했습니다." }, { "label": "D", "score": 3, "verdict_ko": "CCTV 기기에 금지된 텍스트가 생성되었으며, 카메라의 크기가 과장되고 스마트폰을 쥔 손의 해부학적 오류가 큽니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L09B03.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "휴대폰을 쥐고 있는 손의 손가락 마디가 뭉개지고 해부학적으로 기형적입니다.", "fix_en": "Correct the anatomy of the hand holding the phone to show distinct, naturally structured fingers." }, { "issue_ko": "배경에 주차된 자전거의 앞바퀴 아랫부분이 바닥에 녹아내린 듯 형태가 온전하지 않습니다.", "fix_en": "Restore the bicycle's front wheel so that it forms a complete, perfect circle." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Correct the anatomy of the hand holding the phone to show distinct, naturally structured fingers.\n- Restore the bicycle's front wheel so that it forms a complete, perfect circle.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L09B01", "B": "L09B02", "C": "L09B03" }, "assigned": "L09B03", "choice": "Candidate C", "confident": true, "reason_ko": "지문에서 '마트 외부 건물 벽에 달린 CCTV 카메라'를 명시하고 있으며, 후보 C는 마트 외부 벽면에 설치된 CCTV 카메라를 확실하게 보여주므로 가장 적합합니다.", "kept": "L09B03" } }, "S19sh5::variants": { "author_fp": "1a95cbdd9ed7f8b5", "author": { "variants": [ { "approach_ko": "눈높이의 타이트한 측면 클로즈업으로 굳은 표정과 통화 중인 손의 긴장을 포착한다.", "prompt_en": "Frame an intimate eye-level side-profile close-up, cropping tightly around her head, shoulder, and the hand holding the phone. Use a natural portrait perspective with very shallow depth of field, keeping her eye, set jaw, and phone-hand contact critically sharp while the exterior wall falls softly away. Let restrained early-morning daylight rake gently across the profile, emphasizing her rigid stillness and fixed forward gaze." }, { "approach_ko": "멀어진 측면 미디엄 숏과 시선 앞의 넓은 여백으로 불안한 정적과 장소의 압박감을 강조한다.", "prompt_en": "Photograph her in a restrained side-profile medium shot from slightly below eye level, placing her off-center with broad negative space extending in the direction of her stare. Use a compressed perspective and layered facade geometry to hold her firmly against the location rather than isolating her in blur. Keep the early-morning daylight subdued and directional, with her silhouette, phone gesture, backpack, and locked posture reading clearly within the environmental composition." } ] }, "reused": false }, "S19sh5": { "input_fingerprint": "6d71f2123f478d33", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 귀에 핸드폰을 댄 채 굳은 얼굴로 앞을 주시하는 수리영의 측면.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with merchandise aisles and employee areas, plus an exterior bicycle parking area and a facade covered by security cameras. 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 daylight exterior wall, CCTV-camera area, and the young woman's continuing clothing and position from the reference. Exclude the upward-looking pose and the CCTV camera if it falls outside the tighter phone-call framing.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Her bicycle remains parked outside while Su-ri-yeong holds her phone to her ear; the doll-adorned backpack remains 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 귀에 핸드폰을 댄 채 굳은 얼굴로 앞을 주시하는 수리영의 측면.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with merchandise aisles and employee areas, plus an exterior bicycle parking area and a facade covered by security cameras. 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 daylight exterior wall, CCTV-camera area, and the young woman's continuing clothing and position from the reference. Exclude the upward-looking pose and the CCTV camera if it falls outside the tighter phone-call framing.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Her bicycle remains parked outside while Su-ri-yeong holds her phone to her ear; the doll-adorned backpack remains 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame an intimate eye-level side-profile close-up, cropping tightly around her head, shoulder, and the hand holding the phone. Use a natural portrait perspective with very shallow depth of field, keeping her eye, set jaw, and phone-hand contact critically sharp while the exterior wall falls softly away. Let restrained early-morning daylight rake gently across the profile, emphasizing her rigid stillness and fixed forward gaze.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 귀에 핸드폰을 댄 채 굳은 얼굴로 앞을 주시하는 수리영의 측면.\n\nLOCATION (lock): A large supermarket in the quiet early hours, with merchandise aisles and employee areas, plus an exterior bicycle parking area and a facade covered by security cameras. 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 daylight exterior wall, CCTV-camera area, and the young woman's continuing clothing and position from the reference. Exclude the upward-looking pose and the CCTV camera if it falls outside the tighter phone-call framing.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Her bicycle remains parked outside while Su-ri-yeong holds her phone to her ear; the doll-adorned backpack remains 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a restrained side-profile medium shot from slightly below eye level, placing her off-center with broad negative space extending in the direction of her stare. Use a compressed perspective and layered facade geometry to hold her firmly against the location rather than isolating her in blur. Keep the early-morning daylight subdued and directional, with her silhouette, phone gesture, backpack, and locked posture reading clearly within the environmental composition.\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": "측면 구도와 통화 중인 자세를 정확히 연출했으며, 기준 이미지의 건축물 디테일(배전반, 가스 배관 등)을 훌륭하게 유지했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지시된 인물의 행동과 구도는 잘 표현되었으나, 간판과 CCTV의 위치가 기준 이미지의 실제 건축 구조와 다르게 왜곡되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S19sh4_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프롬프트의 '텍스트 금지(No text anywhere)' 규칙에도 불구하고 왼쪽 상단 간판에 글자가 생성되었습니다.", "fix_en": "Remove all text and lettering from the white sign board above the sliding glass doors on the left." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text and lettering from the white sign board above the sliding glass doors on the left.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S19sh4" } }, "S20sh1::variants": { "author_fp": "8ec3ef36d9afdae9", "author": { "variants": [ { "approach_ko": "모니터 바로 옆에서 두 사람의 시선과 긴장을 정면에 가깝게 포착하는 밀착 투샷.", "prompt_en": "Shoot a tight, eye-level chest-up two-shot from beside the CCTV monitor, with the monitor’s dark edge looming softly in the near foreground and both faces seen in three-quarter view beyond it. Use shallow depth of field to hold their concentrated gaze and restrained posture in crisp focus, letting the office recede into subdued softness; the weak screen light shapes their faces against gentle morning ambient illumination." }, { "approach_ko": "두 사람의 뒤쪽 높은 시점에서 어두운 CCTV 화면과 감시 공간의 관계를 강조하는 관찰적 구도.", "prompt_en": "Frame an observational medium shot from slightly above and behind the pair, keeping both upper bodies layered beside the monitoring station while the abnormally dark CCTV image occupies the compositional center. Use deeper focus and restrained perspective so the playback controls, screens, and compact office context remain legible without distracting from the unreadable footage; let the figures form a still, asymmetrical frame around the monitor." } ] }, "reused": false }, "S20sh1": { "input_fingerprint": "fed68c061784491f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 CCTV 모니터 화면을 주시하는 한국인 남성 보안 직원과 그 옆에 선 혜수의 상체.\n\nLOCATION (lock): A supermarket administrative and security office with a CCTV monitoring station, playback controls, and screens used to review recorded exterior footage. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 relevant section of the CCTV recording remains abnormally dark and impossible to identify. Su-ri-yeong still has her doll-adorned backpack and phone 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리); 한국인 남성 보안 직원 (한국인 남성, 성인, 평범한 얼굴, 짧고 단정한 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 CCTV 모니터 화면을 주시하는 한국인 남성 보안 직원과 그 옆에 선 혜수의 상체.\n\nLOCATION (lock): A supermarket administrative and security office with a CCTV monitoring station, playback controls, and screens used to review recorded exterior footage. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 relevant section of the CCTV recording remains abnormally dark and impossible to identify. Su-ri-yeong still has her doll-adorned backpack and phone 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리); 한국인 남성 보안 직원 (한국인 남성, 성인, 평범한 얼굴, 짧고 단정한 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot a tight, eye-level chest-up two-shot from beside the CCTV monitor, with the monitor’s dark edge looming softly in the near foreground and both faces seen in three-quarter view beyond it. Use shallow depth of field to hold their concentrated gaze and restrained posture in crisp focus, letting the office recede into subdued softness; the weak screen light shapes their faces against gentle morning ambient 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 CCTV 모니터 화면을 주시하는 한국인 남성 보안 직원과 그 옆에 선 혜수의 상체.\n\nLOCATION (lock): A supermarket administrative and security office with a CCTV monitoring station, playback controls, and screens used to review recorded exterior footage. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 relevant section of the CCTV recording remains abnormally dark and impossible to identify. Su-ri-yeong still has her doll-adorned backpack and phone 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리); 한국인 남성 보안 직원 (한국인 남성, 성인, 평범한 얼굴, 짧고 단정한 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame an observational medium shot from slightly above and behind the pair, keeping both upper bodies layered beside the monitoring station while the abnormally dark CCTV image occupies the compositional center. Use deeper focus and restrained perspective so the playback controls, screens, and compact office context remain legible without distracting from the unreadable footage; let the figures form a still, asymmetrical frame around the monitor.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 CCTV 모니터 화면을 주시하는 한국인 남성 보안 직원과 그 옆에 선 혜수의 상체.\n\nLOCATION (lock): A supermarket administrative and security office with a CCTV monitoring station, playback controls, and screens used to review recorded exterior footage. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 relevant section of the CCTV recording remains abnormally dark and impossible to identify. Su-ri-yeong still has her doll-adorned backpack and phone 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리); 한국인 남성 보안 직원 (한국인 남성, 성인, 평범한 얼굴, 짧고 단정한 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot a tight, eye-level chest-up two-shot from beside the CCTV monitor, with the monitor’s dark edge looming softly in the near foreground and both faces seen in three-quarter view beyond it. Use shallow depth of field to hold their concentrated gaze and restrained posture in crisp focus, letting the office recede into subdued softness; the weak screen light shapes their faces against gentle morning ambient 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.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): morning, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 어두운 CCTV 모니터 화면을 주시하는 한국인 남성 보안 직원과 그 옆에 선 혜수의 상체.\n\nLOCATION (lock): A supermarket administrative and security office with a CCTV monitoring station, playback controls, and screens used to review recorded exterior footage. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 relevant section of the CCTV recording remains abnormally dark and impossible to identify. Su-ri-yeong still has her doll-adorned backpack and phone 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: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리); 한국인 남성 보안 직원 (한국인 남성, 성인, 평범한 얼굴, 짧고 단정한 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame an observational medium shot from slightly above and behind the pair, keeping both upper bodies layered beside the monitoring station while the abnormally dark CCTV image occupies the compositional center. Use deeper focus and restrained perspective so the playback controls, screens, and compact office context remain legible without distracting from the unreadable footage; let the figures form a still, asymmetrical frame around the monitor.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L15B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S20sh1.png" }, { "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L15B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S20sh1.png" }, { "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L15B01.png" }, { "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L15B01.png" }, { "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", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "상체 위주의 프레이밍이 지시문과 정확히 일치하며, 가죽 재킷과 인형 달린 배낭 등 혜수의 설정이 프레임 내에 완벽히 구현되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "지시된 상체 프레이밍을 벗어나 앵글이 너무 넓으며, 혜수의 의상이 기준 이미지와 완전히 다릅니다." }, { "label": "C", "score": 6, "verdict_ko": "상체 프레이밍과 인물 외형은 훌륭하나, 어깨가 보임에도 배낭 끈이 없어 소지품 지시 반영이 다소 아쉽습니다." }, { "label": "D", "score": 4, "verdict_ko": "지시된 상체보다 넓게 촬영되었고, 혜수의 등이 드러남에도 배낭이 누락되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "상체 위주의 프레이밍이 지시문과 정확히 일치하며, 가죽 재킷과 인형 달린 배낭 등 혜수의 설정이 프레임 내에 완벽히 구현되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "지시된 상체 프레이밍을 벗어나 앵글이 너무 넓으며, 혜수의 의상이 기준 이미지와 완전히 다릅니다." }, { "label": "C", "score": 6, "verdict_ko": "상체 프레이밍과 인물 외형은 훌륭하나, 어깨가 보임에도 배낭 끈이 없어 소지품 지시 반영이 다소 아쉽습니다." }, { "label": "D", "score": 4, "verdict_ko": "지시된 상체보다 넓게 촬영되었고, 혜수의 등이 드러남에도 배낭이 누락되었습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "지정된 혜수의 상체 프레이밍을 정확히 구현했으며, 어두운 모니터를 주시하는 모습과 인형이 달린 배낭, 휴대폰 등 소지품 지시를 훌륭하게 반영했습니다." }, { "label": "B", "score": 6, "verdict_ko": "상체 프레이밍과 인물들의 표정 연기는 우수하나, 프레임 내에 지시된 소지품(배낭과 휴대폰)이 묘사되지 않았습니다." }, { "label": "A", "score": 4, "verdict_ko": "요구된 상체 샷보다 카메라가 너무 물러나 하반신까지 노출되었으며, 필수 소지품이 보이지 않습니다." }, { "label": "C", "score": 3, "verdict_ko": "상체 프레이밍 지시를 완전히 무시한 와이드 샷이며, 배낭을 착용하지 않고 책상에 내려놓아 소지 상태 지시를 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 혜수의 상체 프레이밍을 정확히 구현했으며, 어두운 모니터를 주시하는 모습과 인형이 달린 배낭, 휴대폰 등 소지품 지시를 훌륭하게 반영했습니다." }, { "label": "C", "score": 6, "verdict_ko": "상체 프레이밍과 인물들의 표정 연기는 우수하나, 프레임 내에 지시된 소지품(배낭과 휴대폰)이 묘사되지 않았습니다." }, { "label": "D", "score": 4, "verdict_ko": "요구된 상체 샷보다 카메라가 너무 물러나 하반신까지 노출되었으며, 필수 소지품이 보이지 않습니다." }, { "label": "B", "score": 3, "verdict_ko": "상체 프레이밍 지시를 완전히 무시한 와이드 샷이며, 배낭을 착용하지 않고 책상에 내려놓아 소지 상태 지시를 위반했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 6, "C": 12, "D": 8 }, "ranking": [ "A", "C", "D", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 6, "C": 12, "D": 8 }, "selected": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "상체 위주의 프레이밍이 지시문과 정확히 일치하며, 가죽 재킷과 인형 달린 배낭 등 혜수의 설정이 프레임 내에 완벽히 구현되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "지시된 상체 프레이밍을 벗어나 앵글이 너무 넓으며, 혜수의 의상이 기준 이미지와 완전히 다릅니다." }, { "label": "C", "score": 6, "verdict_ko": "상체 프레이밍과 인물 외형은 훌륭하나, 어깨가 보임에도 배낭 끈이 없어 소지품 지시 반영이 다소 아쉽습니다." }, { "label": "D", "score": 4, "verdict_ko": "지시된 상체보다 넓게 촬영되었고, 혜수의 등이 드러남에도 배낭이 누락되었습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L15B01.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "전경에 위치한 모니터 뒷면에 'LG'라는 텍스트(로고)가 나타나 있습니다.", "fix_en": "Remove the 'LG' text logo from the back of the monitor in the foreground." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the 'LG' text logo from the back of the monitor in the foreground.\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)", "share_plan": { "ref_plan": "background" } }, "S21sh5::variants": { "author_fp": "eae77139fe851316", "author": { "variants": [ { "approach_ko": "정면 눈높이의 밀도 높은 미디엄 클로즈업으로 수리영의 고립감과 정지된 표정을 강조한다.", "prompt_en": "Frame her upper body in a restrained, eye-level medium close-up, held nearly frontal and centered while the surrounding office geometry falls softly out of focus. Use gentle directional interior light across her face with subdued contrast, letting layered desk edges and monitors create a compressed enclosure around her still posture. Keep the visual emphasis on the quiet weight of her solitary presence." }, { "approach_ko": "높은 사선 시점과 깊은 초점으로 복잡한 사무실 속에 홀로 남은 수리영을 작고 고립되게 배치한다.", "prompt_en": "Photograph her upper body from an elevated oblique angle across the clustered workstations, placing her off-center within broad negative space and rigid desk lines. Use a wider, deeper-focus composition so the office structure and the spread photographs remain legible as layered context without competing with her. Cool, even interior illumination and restrained color create a detached, observational mood." } ] }, "reused": false }, "S21sh5": { "input_fingerprint": "40e95a5b82e8501d", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 의자에 홀로 앉아 있는 수리영의 상체.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong remains alone in the office with her doll-adorned backpack and phone, while the enlarged murder-scene photographs stay scattered across Hye-su’s desk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 의자에 홀로 앉아 있는 수리영의 상체.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong remains alone in the office with her doll-adorned backpack and phone, while the enlarged murder-scene photographs stay scattered across Hye-su’s desk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame her upper body in a restrained, eye-level medium close-up, held nearly frontal and centered while the surrounding office geometry falls softly out of focus. Use gentle directional interior light across her face with subdued contrast, letting layered desk edges and monitors create a compressed enclosure around her still posture. Keep the visual emphasis on the quiet weight of her solitary 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.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 의자에 홀로 앉아 있는 수리영의 상체.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong remains alone in the office with her doll-adorned backpack and phone, while the enlarged murder-scene photographs stay scattered across Hye-su’s desk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her upper body from an elevated oblique angle across the clustered workstations, placing her off-center within broad negative space and rigid desk lines. Use a wider, deeper-focus composition so the office structure and the spread photographs remain legible as layered context without competing with her. Cool, even interior illumination and restrained color create a detached, observational mood.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 의자에 홀로 앉아 있는 수리영의 상체.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong remains alone in the office with her doll-adorned backpack and phone, while the enlarged murder-scene photographs stay scattered across Hye-su’s desk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame her upper body in a restrained, eye-level medium close-up, held nearly frontal and centered while the surrounding office geometry falls softly out of focus. Use gentle directional interior light across her face with subdued contrast, letting layered desk edges and monitors create a compressed enclosure around her still posture. Keep the visual emphasis on the quiet weight of her solitary 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.", "D": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 텅 빈 사무실 의자에 홀로 앉아 있는 수리영의 상체.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong remains alone in the office with her doll-adorned backpack and phone, while the enlarged murder-scene photographs stay scattered across Hye-su’s desk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her upper body from an elevated oblique angle across the clustered workstations, placing her off-center within broad negative space and rigid desk lines. Use a wider, deeper-focus composition so the office structure and the spread photographs remain legible as layered context without competing with her. Cool, even interior illumination and restrained color create a detached, observational mood.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S21sh5.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S21sh5.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B02.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", "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "프롬프트에 허락되지 않은 인물들이 배경에 등장하여 치명적인 위반입니다." }, { "label": "B", "score": 5, "verdict_ko": "상체 샷 요구를 무시하고 전신이 포함된 너무 넓은 화각으로 촬영되었습니다." }, { "label": "C", "score": 7, "verdict_ko": "요구된 상체 샷 구도를 완벽히 구현했으며 홀로 있는 상태와 소품들도 정확합니다." }, { "label": "D", "score": 5, "verdict_ko": "상체 샷 지시를 어기고 인물의 전신과 공간이 너무 많이 드러나게 프레이밍되었습니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "D", "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "프롬프트에 허락되지 않은 인물들이 배경에 등장하여 치명적인 위반입니다." }, { "label": "B", "score": 5, "verdict_ko": "상체 샷 요구를 무시하고 전신이 포함된 너무 넓은 화각으로 촬영되었습니다." }, { "label": "C", "score": 7, "verdict_ko": "요구된 상체 샷 구도를 완벽히 구현했으며 홀로 있는 상태와 소품들도 정확합니다." }, { "label": "D", "score": 5, "verdict_ko": "상체 샷 지시를 어기고 인물의 전신과 공간이 너무 많이 드러나게 프레이밍되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A", "C", "D" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "홀로 있는 상황과 소품은 잘 구현했으나, '상체' 샷 지시를 어기고 공간이 넓게 보이는 풀샷으로 연출되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "'상체'라는 정확한 샷 크기를 완벽히 구현했으며, 빈 사무실에 홀로 앉은 상황과 지정된 소품들까지 충실하게 반영했습니다." }, { "label": "C", "score": 4, "verdict_ko": "상체 샷 규정을 무시하고 측면 풀샷을 잡았으며, 불필요한 텍스트(Hye-su)가 노출되어 몰입을 방해합니다." }, { "label": "D", "score": 3, "verdict_ko": "상체 샷은 맞으나, 배경에 명시되지 않은 인물이 등장하여 '텅 빈 사무실에 홀로'라는 핵심 조건을 심각하게 위반했습니다." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "D", "B", "A" ], "verdicts": [ { "label": "D", "score": 5, "verdict_ko": "홀로 있는 상황과 소품은 잘 구현했으나, '상체' 샷 지시를 어기고 공간이 넓게 보이는 풀샷으로 연출되었습니다." }, { "label": "C", "score": 7, "verdict_ko": "'상체'라는 정확한 샷 크기를 완벽히 구현했으며, 빈 사무실에 홀로 앉은 상황과 지정된 소품들까지 충실하게 반영했습니다." }, { "label": "B", "score": 4, "verdict_ko": "상체 샷 규정을 무시하고 측면 풀샷을 잡았으며, 불필요한 텍스트(Hye-su)가 노출되어 몰입을 방해합니다." }, { "label": "A", "score": 3, "verdict_ko": "상체 샷은 맞으나, 배경에 명시되지 않은 인물이 등장하여 '텅 빈 사무실에 홀로'라는 핵심 조건을 심각하게 위반했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 9, "C": 14, "D": 10 }, "ranking": [ "C", "D", "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 9, "C": 14, "D": 10 }, "selected": "C", "ranking": [ "C", "D", "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "프롬프트에 허락되지 않은 인물들이 배경에 등장하여 치명적인 위반입니다." }, { "label": "B", "score": 5, "verdict_ko": "상체 샷 요구를 무시하고 전신이 포함된 너무 넓은 화각으로 촬영되었습니다." }, { "label": "C", "score": 7, "verdict_ko": "요구된 상체 샷 구도를 완벽히 구현했으며 홀로 있는 상태와 소품들도 정확합니다." }, { "label": "D", "score": 5, "verdict_ko": "상체 샷 지시를 어기고 인물의 전신과 공간이 너무 많이 드러나게 프레이밍되었습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: 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": false, "ref_mode": "플레이트+엔티티 (4택1: 무콘티 승·C=변형0)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L14B01", "B": "L14B02", "C": "L14B03" }, "assigned": "L14B02", "choice": "Candidate B", "confident": true, "reason_ko": "주어진 후보 이미지들은 모두 동일한 사무실 공간을 보여주며 서브공간의 차이가 없으므로, 현재 배정된 후보를 유지합니다.", "kept": "L14B02" } }, "S21sh7::variants": { "author_fp": "1390abb45fcedee7", "author": { "variants": [ { "approach_ko": "책상 높이의 정면 밀착 클로즈업으로 수리영의 크게 뜬 눈과 얼어붙은 충격을 고립시킨다.", "prompt_en": "Frame a tight, desk-level close-up from just forward of Su-ri-yeong, catching her face nearly front-on as her gaze drops below the lens toward the photographs. Use very shallow focus so her widened eyes and arrested facial muscles are razor clear while the desk and photograph edges form a soft lower foreground; keep the busy office compressed into subdued blur behind her. Let the established office light fall naturally across her face, preserving realistic skin texture and restrained live-action shock." }, { "approach_ko": "높은 사선의 근접 구도로 사진과 얼굴의 시선을 한 축에 놓아 충격의 원인을 공간적으로 강조한다.", "prompt_en": "Photograph the moment in a high, oblique close-up over the desk, placing the spread photographs in the near foreground and Su-ri-yeong’s shocked face just beyond them as she looks sharply downward. Hold enough depth to retain the physical relationship between her gaze and the photographs while allowing the surrounding office to fall away at the frame edges. Use the existing daytime office illumination with gentle directional contrast, emphasizing her wide eyes and sudden stillness without stylization or anatomical exaggeration." } ] }, "reused": false }, "S21sh7": { "input_fingerprint": "7fa41abab3738634", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 책상 위 사진을 내려다보며 눈을 크게 뜬 수리영의 충격받은 얼굴 클로즈업.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. 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 daytime office lighting, desk arrangement, institutional materials, and the seated woman's clothing from the reference. Exclude the empty-office wide framing and unrelated furniture, focusing on her shocked face as she looks down at the crime photographs.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 gruesome murder-scene photographs remain spread over Hye-su’s desk as Su-ri-yeong examines them; her backpack and phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 책상 위 사진을 내려다보며 눈을 크게 뜬 수리영의 충격받은 얼굴 클로즈업.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. 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 daytime office lighting, desk arrangement, institutional materials, and the seated woman's clothing from the reference. Exclude the empty-office wide framing and unrelated furniture, focusing on her shocked face as she looks down at the crime photographs.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 gruesome murder-scene photographs remain spread over Hye-su’s desk as Su-ri-yeong examines them; her backpack and phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight, desk-level close-up from just forward of Su-ri-yeong, catching her face nearly front-on as her gaze drops below the lens toward the photographs. Use very shallow focus so her widened eyes and arrested facial muscles are razor clear while the desk and photograph edges form a soft lower foreground; keep the busy office compressed into subdued blur behind her. Let the established office light fall naturally across her face, preserving realistic skin texture and restrained live-action shock.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 책상 위 사진을 내려다보며 눈을 크게 뜬 수리영의 충격받은 얼굴 클로즈업.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. 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 daytime office lighting, desk arrangement, institutional materials, and the seated woman's clothing from the reference. Exclude the empty-office wide framing and unrelated furniture, focusing on her shocked face as she looks down at the crime photographs.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 gruesome murder-scene photographs remain spread over Hye-su’s desk as Su-ri-yeong examines them; her backpack and phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in a high, oblique close-up over the desk, placing the spread photographs in the near foreground and Su-ri-yeong’s shocked face just beyond them as she looks sharply downward. Hold enough depth to retain the physical relationship between her gaze and the photographs while allowing the surrounding office to fall away at the frame edges. Use the existing daytime office illumination with gentle directional contrast, emphasizing her wide eyes and sudden stillness without stylization or anatomical exaggeration.\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": "지시된 '얼굴 클로즈업' 샷 크기를 완벽하게 구현했으며, 내려다보는 충격받은 표정도 잘 표현되었습니다." }, { "label": "B", "score": 5, "verdict_ko": "표정과 소품은 좋으나, 프레임이 너무 넓어 요구된 '얼굴 클로즈업' 대신 미디엄 샷에 가깝게 연출되었습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S21sh5_sel.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 흐릿한 인물들이 배경에 추가됨.", "fix_en": "Remove all blurred figures of other people from the background." }, { "issue_ko": "전경에 놓인 사진들이 캐릭터가 아닌 카메라(시청자)를 향해 정방향으로 배치되어 있음.", "fix_en": "Rotate the foreground photographs 180 degrees so they face the character looking down at them, not the camera." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all blurred figures of other people from the background.\n- Rotate the foreground photographs 180 degrees so they face the character looking down at them, not the camera.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S21sh5" } }, "S21sh8::variants": { "author_fp": "bd47bc7bf7de2909", "author": { "variants": [ { "approach_ko": "모니터 화면을 정면에서 밀착해 좌표와 오프라인 상태를 냉정하게 강조하는 정지 인서트.", "prompt_en": "A locked-off, tight frontal insert centered on the illuminated monitor, with minimal perspective distortion and a precise crop that makes the maritime map, coordinates, and offline status the unmistakable visual focus. Keep the screen plane crisp and evenly resolved while the surrounding workstation falls softly out of focus, creating a clinical, ominous stillness." }, { "approach_ko": "책상 높이의 비스듬한 중근경으로 모니터와 수사 공간의 깊이를 함께 담는 환경적 구도.", "prompt_en": "Photograph from an oblique desk-height angle in a wider environmental composition, placing the active monitor prominently within the foreground workstation while the clustered office recedes in layered depth. Hold enough focus on the screen for its critical display to remain clear, while the nearby desk photographs and softly rendered police-office background frame the information as part of an ongoing investigation; use restrained naturalistic contrast and a static, observational mood." } ] }, "reused": false }, "S21sh8": { "input_fingerprint": "0bb46f8b5c1ec9d7", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 켜진 컴퓨터 모니터 화면 속 해상 지도 위 좌표와 오프라인 상태 메시지가 뚜렷한 정지 구도.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 monitor continues displaying Minsuk’s last offshore coordinates and an offline status. Su-ri-yeong still has her backpack and phone, with the desk photographs remaining nearby.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 켜진 컴퓨터 모니터 화면 속 해상 지도 위 좌표와 오프라인 상태 메시지가 뚜렷한 정지 구도.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 monitor continues displaying Minsuk’s last offshore coordinates and an offline status. Su-ri-yeong still has her backpack and phone, with the desk photographs remaining nearby.\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):\nA locked-off, tight frontal insert centered on the illuminated monitor, with minimal perspective distortion and a precise crop that makes the maritime map, coordinates, and offline status the unmistakable visual focus. Keep the screen plane crisp and evenly resolved while the surrounding workstation falls softly out of focus, creating a clinical, ominous 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 켜진 컴퓨터 모니터 화면 속 해상 지도 위 좌표와 오프라인 상태 메시지가 뚜렷한 정지 구도.\n\nLOCATION (lock): A busy police station office with clustered desks, computer monitors, paperwork, and adjoining access to a meeting area. Enlarged crime-scene photographs are spread across one investigator's desk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 monitor continues displaying Minsuk’s last offshore coordinates and an offline status. Su-ri-yeong still has her backpack and phone, with the desk photographs remaining nearby.\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 desk-height angle in a wider environmental composition, placing the active monitor prominently within the foreground workstation while the clustered office recedes in layered depth. Hold enough focus on the screen for its critical display to remain clear, while the nearby desk photographs and softly rendered police-office background frame the information as part of an ongoing investigation; use restrained naturalistic contrast and a static, observational mood.\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": 4 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "해상 지도 위의 오프라인 메시지를 명확하게 묘사했으며, 배낭과 현장 사진 등 주변 소품과 지정된 배경 공간을 충실히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "모니터 화면 구도는 좋으나, 요구된 해상 지도가 아닌 내륙 지도를 표시하여 프롬프트의 핵심 내용을 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L14B02.png" } ], "critique": { "issues": [ { "issue_ko": "수리영이 배낭을 소지하고 있다는 설정(still has her backpack)과 다르게, 프레임에 없는 그녀의 배낭이 책상 위에 남겨져 있습니다.", "fix_en": "Remove the backpack from the desk completely." }, { "issue_ko": "텍스트 생성 금지(No text anywhere) 규칙이 있음에도 불구하고 모니터 화면 중앙에 '오프라인', 'OFFLINE', '신호 없음' 등의 문자가 직접적으로 표시되어 있습니다.", "fix_en": "Remove all written text from the monitor screen, representing the offline status message using only a graphical symbol or warning icon." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the backpack from the desk completely.\n- Remove all written text from the monitor screen, representing the offline status message using only a graphical symbol or warning icon.\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)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S21sh7" }, "plate_select": { "candidates": { "A": "L14B01", "B": "L14B02", "C": "L14B03" }, "assigned": "L14B01", "choice": "Candidate A", "confident": true, "reason_ko": "모든 후보가 동일한 사무실 공간(서브공간)을 보여주고 있으며, 샷에서 요구하는 '컴퓨터 모니터'가 공통적으로 존재하므로 서브공간을 변경할 명확한 근거가 없어 현재 할당된 후보를 유지합니다.", "kept": "L14B01" } }, "S22sh2::variants": { "author_fp": "cb7c8b191dfdfc46", "author": { "variants": [ { "approach_ko": "안장 높이의 측면 트래킹으로 수리영과 자전거를 선명하게 붙잡고 배경의 흐름으로 속도를 강조한다.", "prompt_en": "Photograph the moment in a close side-profile tracking view from roughly saddle height, moving parallel with Su-ri-yeong so her clenched hands, deeply folded posture, active pedal stroke, and backward-streaming hair and clothing remain sharply legible. Use a moderately compressed lens feel and layered lateral background blur to convey urgent forward speed, with clear lead room ahead of her. Keep the existing daylight natural and grounded in the location." }, { "approach_ko": "낮고 넓은 고정 시점에서 수리영이 프레임을 가로지르게 하여 공간과 속도감을 동시에 드러낸다.", "prompt_en": "Use a low, wide, fixed side-on viewpoint that lets Su-ri-yeong drive across the police-station entrance and parking space within a strong environmental composition. Hold her full riding silhouette and bicycle geometry clearly against the deeper setting, catching the pedal mid-cycle while a restrained trace of motion softens her flying hair and clothing without obscuring her grip or posture. Let the location’s natural daytime light and depth carry the realism, with open space in her direction of travel." } ] }, "reused": false }, "S22sh2": { "input_fingerprint": "ea4730d71e327105", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 핸들을 꽉 쥐고 허리를 한껏 숙인 채 페달을 밟으며 앞으로 나아가는 중간 순간의 수리영 측면, 머리카락과 옷자락이 움직임의 반대 방향으로 흩날리는 모습.\n\nLOCATION (lock): The exterior entrance and parking area of a police station, with a bicycle parked nearby and direct access to the street. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong rides her bicycle with the folded coordinate note in her pocket, while her doll-adorned backpack and mobile phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 핸들을 꽉 쥐고 허리를 한껏 숙인 채 페달을 밟으며 앞으로 나아가는 중간 순간의 수리영 측면, 머리카락과 옷자락이 움직임의 반대 방향으로 흩날리는 모습.\n\nLOCATION (lock): The exterior entrance and parking area of a police station, with a bicycle parked nearby and direct access to the street. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong rides her bicycle with the folded coordinate note in her pocket, while her doll-adorned backpack and mobile phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in a close side-profile tracking view from roughly saddle height, moving parallel with Su-ri-yeong so her clenched hands, deeply folded posture, active pedal stroke, and backward-streaming hair and clothing remain sharply legible. Use a moderately compressed lens feel and layered lateral background blur to convey urgent forward speed, with clear lead room ahead of her. Keep the existing daylight natural and grounded in the location.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 핸들을 꽉 쥐고 허리를 한껏 숙인 채 페달을 밟으며 앞으로 나아가는 중간 순간의 수리영 측면, 머리카락과 옷자락이 움직임의 반대 방향으로 흩날리는 모습.\n\nLOCATION (lock): The exterior entrance and parking area of a police station, with a bicycle parked nearby and direct access to the street. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong rides her bicycle with the folded coordinate note in her pocket, while her doll-adorned backpack and mobile phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a low, wide, fixed side-on viewpoint that lets Su-ri-yeong drive across the police-station entrance and parking space within a strong environmental composition. Hold her full riding silhouette and bicycle geometry clearly against the deeper setting, catching the pedal mid-cycle while a restrained trace of motion softens her flying hair and clothing without obscuring her grip or posture. Let the location’s natural daytime light and depth carry the realism, with open space in her direction of travel.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 핸들을 꽉 쥐고 허리를 한껏 숙인 채 페달을 밟으며 앞으로 나아가는 중간 순간의 수리영 측면, 머리카락과 옷자락이 움직임의 반대 방향으로 흩날리는 모습.\n\nLOCATION (lock): The exterior entrance and parking area of a police station, with a bicycle parked nearby and direct access to the street. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong rides her bicycle with the folded coordinate note in her pocket, while her doll-adorned backpack and mobile phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in a close side-profile tracking view from roughly saddle height, moving parallel with Su-ri-yeong so her clenched hands, deeply folded posture, active pedal stroke, and backward-streaming hair and clothing remain sharply legible. Use a moderately compressed lens feel and layered lateral background blur to convey urgent forward speed, with clear lead room ahead of her. Keep the existing daylight natural and grounded in the location.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 핸들을 꽉 쥐고 허리를 한껏 숙인 채 페달을 밟으며 앞으로 나아가는 중간 순간의 수리영 측면, 머리카락과 옷자락이 움직임의 반대 방향으로 흩날리는 모습.\n\nLOCATION (lock): The exterior entrance and parking area of a police station, with a bicycle parked nearby and direct access to the street. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong rides her bicycle with the folded coordinate note in her pocket, while her doll-adorned backpack and mobile phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a low, wide, fixed side-on viewpoint that lets Su-ri-yeong drive across the police-station entrance and parking space within a strong environmental composition. Hold her full riding silhouette and bicycle geometry clearly against the deeper setting, catching the pedal mid-cycle while a restrained trace of motion softens her flying hair and clothing without obscuring her grip or posture. Let the location’s natural daytime light and depth carry the realism, with open space in her direction of travel.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L16B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S22sh2.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L16B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S22sh2.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L16B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L16B01.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", "A", "C", "D" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지정된 자세는 양호하게 표현되었으나 주머니의 쪽지와 배경의 자전거가 생략되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "요구된 역동적 측면 자세, 흩날리는 머리카락, 소지품(쪽지 포함), 배경을 가장 정확히 재현했습니다." }, { "label": "C", "score": 5, "verdict_ko": "지정된 쪽지가 누락되었으며 자전거 핸들을 쥔 손의 구조가 어색합니다." }, { "label": "D", "score": 3, "verdict_ko": "경찰서 간판 텍스트에 심각한 철자 오류('POIJCE')가 발생하여 감점되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A", "C", "D" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지정된 자세는 양호하게 표현되었으나 주머니의 쪽지와 배경의 자전거가 생략되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "요구된 역동적 측면 자세, 흩날리는 머리카락, 소지품(쪽지 포함), 배경을 가장 정확히 재현했습니다." }, { "label": "C", "score": 5, "verdict_ko": "지정된 쪽지가 누락되었으며 자전거 핸들을 쥔 손의 구조가 어색합니다." }, { "label": "D", "score": 3, "verdict_ko": "경찰서 간판 텍스트에 심각한 철자 오류('POIJCE')가 발생하여 감점되었습니다." } ] }, "reverse_raw": { "winner": "C", "ranking": [ "C", "B", "A", "D" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "역동적인 포즈는 좋으나 간판 텍스트(POIJCE) 변형 오류가 있고 주머니의 메모지가 누락됨." }, { "label": "B", "score": 6, "verdict_ko": "동작 표현과 배경은 무난하나, 자전거 앞바퀴 구조의 물리적 오류가 있으며 메모지가 보이지 않음." }, { "label": "C", "score": 8, "verdict_ko": "요구된 측면 구도와 역동적인 포즈를 잘 구현했으며, 메모지를 포함한 소지품과 배경 묘사가 가장 정확함." }, { "label": "D", "score": 4, "verdict_ko": "자전거 앞바퀴와 포크가 완전히 분리된 심각한 물리적 형태 오류가 발생함." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "C", "D", "A" ], "verdicts": [ { "label": "D", "score": 5, "verdict_ko": "역동적인 포즈는 좋으나 간판 텍스트(POIJCE) 변형 오류가 있고 주머니의 메모지가 누락됨." }, { "label": "C", "score": 6, "verdict_ko": "동작 표현과 배경은 무난하나, 자전거 앞바퀴 구조의 물리적 오류가 있으며 메모지가 보이지 않음." }, { "label": "B", "score": 8, "verdict_ko": "요구된 측면 구도와 역동적인 포즈를 잘 구현했으며, 메모지를 포함한 소지품과 배경 묘사가 가장 정확함." }, { "label": "A", "score": 4, "verdict_ko": "자전거 앞바퀴와 포크가 완전히 분리된 심각한 물리적 형태 오류가 발생함." } ] }, "combined": { "totals": { "A": 10, "B": 15, "C": 11, "D": 8 }, "ranking": [ "B", "C", "A", "D" ], "agreement": true, "policy": 1 } }, "totals": { "A": 10, "B": 15, "C": 11, "D": 8 }, "selected": "B", "ranking": [ "B", "C", "A", "D" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "지정된 자세는 양호하게 표현되었으나 주머니의 쪽지와 배경의 자전거가 생략되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "요구된 역동적 측면 자세, 흩날리는 머리카락, 소지품(쪽지 포함), 배경을 가장 정확히 재현했습니다." }, { "label": "C", "score": 5, "verdict_ko": "지정된 쪽지가 누락되었으며 자전거 핸들을 쥔 손의 구조가 어색합니다." }, { "label": "D", "score": 3, "verdict_ko": "경찰서 간판 텍스트에 심각한 철자 오류('POIJCE')가 발생하여 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L16B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "핸들을 쥐고 있는 수리영의 오른손 손가락 구조가 무너져 자전거 브레이크 및 손잡이와 융합되어 있음.", "fix_en": "Restore the anatomy of the right hand so the fingers are distinctly formed and naturally wrapping around the handlebar." }, { "issue_ko": "'텍스트 없음' 지시에도 불구하고 경찰서 건물 간판과 유리문에 형태가 왜곡된 문자들이 생성되어 있음.", "fix_en": "Remove all text from the building's signs and glass doors to strictly comply with the no-text rule." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the anatomy of the right hand so the fingers are distinctly formed and naturally wrapping around the handlebar.\n- Remove all text from the building's signs and glass doors to strictly comply with the no-text rule.\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)", "share_plan": { "ref_plan": "background" } }, "S23sh1::variants": { "author_fp": "544f9493f57f38fd", "author": { "variants": [ { "approach_ko": "모니터 가장자리를 전경에 둔 타이트한 측면 클로즈업으로 수리영의 집중된 시선과 미세한 표정을 강조한다.", "prompt_en": "Photograph a tight side-profile upper-body close-up from seated eye level, with the near edge of the monitor forming a soft foreground plane and her face held in crisp focus. Use shallow depth and restrained monitor spill across her features, emphasizing the unwavering line of her gaze and the minute tension in her posture while the café recedes unobtrusively." }, { "approach_ko": "측면 중경의 깊은 구도로 수리영과 모니터를 한 화면에 분리 배치해 PC방 공간 속 고립감과 지도에 대한 몰입을 보여준다.", "prompt_en": "Use a wider lateral medium shot at monitor height, composing her upper body and the monitor as distinct opposing masses across the frame. Keep more of the computer-station depth legible with layered perspective and moderate depth of field, using subdued interior illumination and the screen as a localized visual anchor to make her concentrated stillness feel isolated within the café." } ] }, "reused": false }, "S23sh1": { "input_fingerprint": "dc677514360aafdc", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): PC방 모니터 화면의 지도를 주시하는 수리영의 측면 상체.\n\nLOCATION (lock): A public computer café with individual monitor stations, internet-connected computers, and access to a printer for producing paper maps. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong keeps the coordinate note and phone with her and places the folded printed maps inside 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): PC방 모니터 화면의 지도를 주시하는 수리영의 측면 상체.\n\nLOCATION (lock): A public computer café with individual monitor stations, internet-connected computers, and access to a printer for producing paper maps. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong keeps the coordinate note and phone with her and places the folded printed maps inside 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph a tight side-profile upper-body close-up from seated eye level, with the near edge of the monitor forming a soft foreground plane and her face held in crisp focus. Use shallow depth and restrained monitor spill across her features, emphasizing the unwavering line of her gaze and the minute tension in her posture while the café recedes unobtrusively.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): PC방 모니터 화면의 지도를 주시하는 수리영의 측면 상체.\n\nLOCATION (lock): A public computer café with individual monitor stations, internet-connected computers, and access to a printer for producing paper maps. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong keeps the coordinate note and phone with her and places the folded printed maps inside 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider lateral medium shot at monitor height, composing her upper body and the monitor as distinct opposing masses across the frame. Keep more of the computer-station depth legible with layered perspective and moderate depth of field, using subdued interior illumination and the screen as a localized visual anchor to make her concentrated stillness feel isolated within the café.\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": "측면 앵글과 캐릭터 외모를 정확히 구현했으나, 지도가 가방 안이 아닌 밖에 놓여 있어 일부 감점되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "캐릭터의 외모가 레퍼런스와 일치하지 않으며, 지시된 소지품(휴대폰, 메모장)이 프레임에 나타나지 않습니다." } ], "refs": [ { "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 인물(왼쪽 배경의 남성)이 등장하여 인물 추가 금지 규칙을 위반했습니다.", "fix_en": "Remove the blurred man sitting at the computer in the background on the left." }, { "issue_ko": "모니터 화면, 배낭 위의 서류, 키보드 옆의 메모지, 배경 벽면의 안내문 등에 텍스트가 존재하여 텍스트 금지 규칙을 위반했습니다.", "fix_en": "Remove all text from the monitor screen, the documents on the backpack, the sticky note by the keyboard, and the wall signs in the background." }, { "issue_ko": "접힌 지도를 배낭 안에 넣는다는 설정과 달리, 서류 뭉치가 배낭 위에 올려져 있습니다.", "fix_en": "Remove the stack of papers resting on top of the backpack." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the blurred man sitting at the computer in the background on the left.\n- Remove all text from the monitor screen, the documents on the backpack, the sticky note by the keyboard, and the wall signs in the background.\n- Remove the stack of papers resting on top of the backpack.\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", "share_plan": { "ref_plan": "background" } }, "S24sh3::variants": { "author_fp": "3b5448958560f836", "author": { "variants": [ { "approach_ko": "정면 눈높이의 밀착 클로즈업으로 굳은 시선과 빗물에 젖은 얼굴을 정면에서 압박감 있게 포착한다.", "prompt_en": "Use an extremely tight, eye-level frontal close-up, holding her eyes near the upper third while the lower edge crops firmly below the chin. Keep both eyes sharply resolved and let the coastal route and sea dissolve into a soft, storm-darkened background; cold diffuse late-afternoon skylight and rain across her face emphasize the rigid stillness of her expression." }, { "approach_ko": "자전거와 나란히 움직이는 낮은 측면 트래킹 클로즈업으로 전진하는 긴장과 앞을 겨누는 시선을 강조한다.", "prompt_en": "Photograph her in a low, close three-quarter profile from alongside her direction of travel, with her face filling the frame and her fixed forward gaze aimed into compressed negative space ahead. Use a parallel tracking feel: keep her features crisp while the rain-lashed coastal surroundings trail into restrained directional blur, shaped by soft storm-cloud light with subtle contour along the wet cheek and jaw." } ] }, "reused": false }, "S24sh3": { "input_fingerprint": "77513b8872432378", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late afternoon, heavy rain and thunderclouds.\n\nSHOT TEXT (authoritative, Korean): 비에 흠뻑 젖은 채 굳은 표정으로 앞을 노려보는 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): An exposed coastal route beside the sea, traversed by bicycle during a heavy shower beneath dense storm clouds. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong continues riding her bicycle with the printed maps secured in her backpack, while the coordinate note and phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late afternoon, heavy rain and thunderclouds.\n\nSHOT TEXT (authoritative, Korean): 비에 흠뻑 젖은 채 굳은 표정으로 앞을 노려보는 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): An exposed coastal route beside the sea, traversed by bicycle during a heavy shower beneath dense storm clouds. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong continues riding her bicycle with the printed maps secured in her backpack, while the coordinate note and phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an extremely tight, eye-level frontal close-up, holding her eyes near the upper third while the lower edge crops firmly below the chin. Keep both eyes sharply resolved and let the coastal route and sea dissolve into a soft, storm-darkened background; cold diffuse late-afternoon skylight and rain across her face emphasize 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): late afternoon, heavy rain and thunderclouds.\n\nSHOT TEXT (authoritative, Korean): 비에 흠뻑 젖은 채 굳은 표정으로 앞을 노려보는 수리영의 얼굴 클로즈업.\n\nLOCATION (lock): An exposed coastal route beside the sea, traversed by bicycle during a heavy shower beneath dense storm clouds. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong continues riding her bicycle with the printed maps secured in her backpack, while the coordinate note and phone remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in a low, close three-quarter profile from alongside her direction of travel, with her face filling the frame and her fixed forward gaze aimed into compressed negative space ahead. Use a parallel tracking feel: keep her features crisp while the rain-lashed coastal surroundings trail into restrained directional blur, shaped by soft storm-cloud light with subtle contour along the wet cheek and jaw.\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": "요청된 '얼굴 클로즈업' 프레이밍과 정면을 노려보는 시선 처리를 정확히 구현하여 가장 높은 우선순위의 지시를 충족했습니다." }, { "label": "B", "score": 4, "verdict_ko": "자전거와 소품 등을 보여주기 위해 프레임을 임의로 넓히면서 명시된 '얼굴 클로즈업' 기준을 위반했습니다." } ], "refs": [ { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "화면 하단이 턱 바로 아래에서 잘리지 않고 어깨와 가슴 윗부분까지 너무 넓게 프레이밍되었습니다.", "fix_en": "Crop the bottom edge of the frame tighter so it cuts off firmly just below the chin, removing the shoulders from view." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Crop the bottom edge of the frame tighter so it cuts off firmly just below the chin, removing the shoulders from view.\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", "share_plan": { "ref_plan": "background" } }, "S25sh3::variants": { "author_fp": "8793037ac390fe43", "author": { "variants": [ { "approach_ko": "거실의 난장판을 넓은 전경으로 두고 열린 방문을 프레임 속 프레임으로 삼아 붉은 원과 사진을 깊숙이 포착한다.", "prompt_en": "Photograph from a low, moderately wide viewpoint in the living room, using the open doorway as a strong frame within the frame. Let the ransacked foreground occupy substantial space while clear depth lines lead into the bedroom, where the marked wall and its central photograph remain the precise visual destination. Cool dusk light from the open window grazes the disorder and falls away toward the bedroom, creating tense, natural depth without obscuring the key detail." }, { "approach_ko": "방문 가장자리를 전경에 크게 걸고 망원 압축으로 붉은 원과 피 묻은 사진을 불길하게 밀착 강조한다.", "prompt_en": "Use a tight, compressed view through the open doorway, with the near door edge occupying one side as a soft, partial foreground obstruction. Flatten the depth so the rough wall mark dominates the composition and the photograph at its center becomes the sharpest, most immediate detail; allow only restrained peripheral evidence of the surrounding disorder. Keep the dusk interior subdued and directional, emphasizing surface texture, dried staining, and the unsettling stillness of the room." } ] }, "reused": false }, "S25sh3": { "input_fingerprint": "c51330ce2aad653f", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 열린 방문 너머, 벽면에 거칠게 그려진 붉은색 원과 그 중앙에 꽂힌 피 묻은 낡은 사진이 보이는 구도.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): The room remains ransacked, with footprints, an open window, and overturned chairs. A bloodstained old photograph is pinned at the center of the red wall circle, while Su-ri-yeong still has her backpack, maps, note, and phone.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 열린 방문 너머, 벽면에 거칠게 그려진 붉은색 원과 그 중앙에 꽂힌 피 묻은 낡은 사진이 보이는 구도.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): The room remains ransacked, with footprints, an open window, and overturned chairs. A bloodstained old photograph is pinned at the center of the red wall circle, while Su-ri-yeong still has her backpack, maps, note, and phone.\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, moderately wide viewpoint in the living room, using the open doorway as a strong frame within the frame. Let the ransacked foreground occupy substantial space while clear depth lines lead into the bedroom, where the marked wall and its central photograph remain the precise visual destination. Cool dusk light from the open window grazes the disorder and falls away toward the bedroom, creating tense, natural depth without obscuring the key 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.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 열린 방문 너머, 벽면에 거칠게 그려진 붉은색 원과 그 중앙에 꽂힌 피 묻은 낡은 사진이 보이는 구도.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): The room remains ransacked, with footprints, an open window, and overturned chairs. A bloodstained old photograph is pinned at the center of the red wall circle, while Su-ri-yeong still has her backpack, maps, note, and phone.\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 tight, compressed view through the open doorway, with the near door edge occupying one side as a soft, partial foreground obstruction. Flatten the depth so the rough wall mark dominates the composition and the photograph at its center becomes the sharpest, most immediate detail; allow only restrained peripheral evidence of the surrounding disorder. Keep the dusk interior subdued and directional, emphasizing surface texture, dried staining, and the unsettling stillness of the room.\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": 2 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "방문 너머 침실 벽면에 그려진 붉은 원과 사진을 사실적으로 구현했으며, 난장판이 된 실내와 발자국 등 지시사항을 잘 따랐습니다." }, { "label": "B", "score": 2, "verdict_ko": "벽면에 있어야 할 붉은 원과 사진이 문틈 허공(또는 렌즈 앞)에 떠 있어 물리적으로 불가능한 연출이라는 치명적인 오류가 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.png" } ], "critique": { "issues": [ { "issue_ko": "거실 창문이 열려 있어야 하지만 닫혀 있습니다.", "fix_en": "Open the closed window on the right side of the room." }, { "issue_ko": "레퍼런스 사진에 있는 열린 나무 방문(문짝)이 완전히 누락되었고, 문틀 바닥에 부자연스러운 높은 턱이 생겼습니다.", "fix_en": "Add the open solid wooden door to the doorframe and remove the raised wooden sill from the floor threshold." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Open the closed window on the right side of the room.\n- Add the open solid wooden door to the doorframe and remove the raised wooden sill from the floor threshold.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B04", "choice": "Candidate A", "confident": true, "reason_ko": "\"열린 방문 너머, 벽면에 거칠게 그려진 붉은색 원\"이라는 묘사에 일치하게 붉은 원이 그려진 벽이 보이는 유일한 이미지입니다.", "kept": "L05B04" } }, "S25sh5::variants": { "author_fp": "c8fae06930223de5", "author": { "variants": [ { "approach_ko": "사진 정면을 거의 화면 가득 담아 두 소녀와 식당의 단서를 선명하게 읽히게 하는 기록물식 클로즈업.", "prompt_en": "Photograph the faded image nearly straight-on in an extremely tight, centered close-up, with its pictured subjects and architectural detail immediately readable. Keep the photograph’s plane largely sharp while the ransacked bedroom falls into complete, unobtrusive blur beyond it; soft dusk light from the open window reveals the faded emulsion, worn surface, and blood staining with restrained forensic clarity." }, { "approach_ko": "사진을 비스듬한 초근접 시점과 얕은 초점으로 훑어 낡은 표면과 불길한 단서를 촉각적으로 강조한다.", "prompt_en": "Use an intimate oblique close-up that looks across the photograph rather than squarely at it, creating strong near-to-far depth within the print. Let shallow focus settle on the two girls and the round roof ornament while the photograph’s edges and distant portions soften naturally; low, grazing dusk illumination from the open window catches creases, faded texture, and bloodstained areas, giving the clue a tactile and ominous presence." } ] }, "reused": false }, "S25sh5": { "input_fingerprint": "df6c5eb4421a4943", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 두 명의 어린 한국인 여자아이와 지붕 위 둥근 원 장식이 있는 낡은 시골 식당이 찍힌 빛바랜 사진의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): Su-ri-yeong has removed the bloodstained photograph from the red circle and holds it for inspection; the ransacked room and the now-empty red wall circle remain behind her. Her backpack still contains the printed maps.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 두 명의 어린 한국인 여자아이와 지붕 위 둥근 원 장식이 있는 낡은 시골 식당이 찍힌 빛바랜 사진의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): Su-ri-yeong has removed the bloodstained photograph from the red circle and holds it for inspection; the ransacked room and the now-empty red wall circle remain behind her. Her backpack still contains the printed maps.\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 faded image nearly straight-on in an extremely tight, centered close-up, with its pictured subjects and architectural detail immediately readable. Keep the photograph’s plane largely sharp while the ransacked bedroom falls into complete, unobtrusive blur beyond it; soft dusk light from the open window reveals the faded emulsion, worn surface, and blood staining with restrained forensic clarity.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 두 명의 어린 한국인 여자아이와 지붕 위 둥근 원 장식이 있는 낡은 시골 식당이 찍힌 빛바랜 사진의 클로즈업.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): Su-ri-yeong has removed the bloodstained photograph from the red circle and holds it for inspection; the ransacked room and the now-empty red wall circle remain behind her. Her backpack still contains the printed maps.\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 an intimate oblique close-up that looks across the photograph rather than squarely at it, creating strong near-to-far depth within the print. Let shallow focus settle on the two girls and the round roof ornament while the photograph’s edges and distant portions soften naturally; low, grazing dusk illumination from the open window catches creases, faded texture, and bloodstained areas, giving the clue a tactile and ominous 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": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 클로즈업 샷을 정확히 구현했으며, 낡은 사진 속 피사체(소녀들, 둥근 장식 식당)와 배경의 장소 설정(침대, 붉은 원, 어질러진 방) 모두 지시사항에 매우 충실합니다." }, { "label": "B", "score": 3, "verdict_ko": "사진이 허공에 떠 있는 물리적으로 불가능한 연출(Hard Violation)을 보였으며, 배경의 공간 구조도 레퍼런스와 다소 어긋납니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B07.png" } ], "critique": { "issues": [ { "issue_ko": "사진 속 건물 창문과 간판에 '식당'이라는 텍스트가 포함되어 있습니다.", "fix_en": "Remove all text from the building in the photograph." }, { "issue_ko": "프롬프트에 신체 부위가 나타나지 않아야 한다고 명시되어 있으나, 화면 좌측 하단에 손가락이 보입니다.", "fix_en": "Remove the fingers visible in the bottom left corner." }, { "issue_ko": "열린 창문을 요구했으나, 배경에 보이는 창문이 완전히 닫혀 있습니다.", "fix_en": "Slide the window panes open in the blurred background." } ] }, "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 building in the photograph.\n- Remove the fingers visible in the bottom left corner.\n- Slide the window panes open in the blurred background.\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)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S25sh3" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B04", "choice": "Candidate D", "confident": true, "reason_ko": "원문에서 언급된 사진은 로케이션 설명 상 침실(bedroom)의 붉은 원이 표시된 벽에 위치해 있으므로, 침실 공간을 보여주는 현재 할당된 이미지를 유지하는 것이 적합합니다.", "kept": "L05B04" } }, "S25sh7::variants": { "author_fp": "d31828f88ae46841", "author": { "variants": [ { "approach_ko": "방 안쪽의 눈높이 와이드 숏으로 폐허가 된 공간의 깊이 끝에 현관 앞 수리영을 작고 고립되게 배치한다.", "prompt_en": "Photograph from deep inside the apartment at standing eye level in a restrained wide composition, placing Su-ri-yeong near the far end of the room at the open entrance. Use the ransacked interior as layered foreground and midground depth, with converging room lines drawing toward her rigid back; keep the backpack, doll adornment, and clenched photograph clearly legible. Let soft dusk spill from the open window rake gently across the disorder while her figure remains naturally subdued against the doorway." }, { "approach_ko": "손 높이의 밀착된 후면 숏으로 백팩과 구겨진 사진을 움켜쥔 손의 긴장을 중심에 둔다.", "prompt_en": "Frame a close rear view from slightly below waist height, compressing the doorway around Su-ri-yeong and making the tension in her gripping hand the visual anchor. Keep her backpack and doll adornment prominent while the crumpled photograph is seen naturally from the camera-facing side dictated by her grip; allow only selective fragments of the disturbed apartment to remain behind her in shallow focus. Use low, cool dusk ambience with a soft edge from the open window, preserving tactile detail in her hand, the worn photograph, and the backpack." } ] }, "reused": false }, "S25sh7": { "input_fingerprint": "8871ef3495477a82", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 백팩을 메고 구겨진 사진을 손에 꽉 쥔 채 열린 현관문 앞에 선 수리영의 뒷모습.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): Su-ri-yeong leaves wearing her doll-adorned backpack and tightly gripping the old photograph; the printed maps remain inside the pack and her phone remains with her. The ransacked room and red circle are left intact behind 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 백팩을 메고 구겨진 사진을 손에 꽉 쥔 채 열린 현관문 앞에 선 수리영의 뒷모습.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): Su-ri-yeong leaves wearing her doll-adorned backpack and tightly gripping the old photograph; the printed maps remain inside the pack and her phone remains with her. The ransacked room and red circle are left intact behind 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from deep inside the apartment at standing eye level in a restrained wide composition, placing Su-ri-yeong near the far end of the room at the open entrance. Use the ransacked interior as layered foreground and midground depth, with converging room lines drawing toward her rigid back; keep the backpack, doll adornment, and clenched photograph clearly legible. Let soft dusk spill from the open window rake gently across the disorder while her figure remains naturally subdued against the doorway.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 백팩을 메고 구겨진 사진을 손에 꽉 쥔 채 열린 현관문 앞에 선 수리영의 뒷모습.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): Su-ri-yeong leaves wearing her doll-adorned backpack and tightly gripping the old photograph; the printed maps remain inside the pack and her phone remains with her. The ransacked room and red circle are left intact behind 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a close rear view from slightly below waist height, compressing the doorway around Su-ri-yeong and making the tension in her gripping hand the visual anchor. Keep her backpack and doll adornment prominent while the crumpled photograph is seen naturally from the camera-facing side dictated by her grip; allow only selective fragments of the disturbed apartment to remain behind her in shallow focus. Use low, cool dusk ambience with a soft edge from the open window, preserving tactile detail in her hand, the worn photograph, and the backpack.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 백팩을 메고 구겨진 사진을 손에 꽉 쥔 채 열린 현관문 앞에 선 수리영의 뒷모습.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): Su-ri-yeong leaves wearing her doll-adorned backpack and tightly gripping the old photograph; the printed maps remain inside the pack and her phone remains with her. The ransacked room and red circle are left intact behind 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from deep inside the apartment at standing eye level in a restrained wide composition, placing Su-ri-yeong near the far end of the room at the open entrance. Use the ransacked interior as layered foreground and midground depth, with converging room lines drawing toward her rigid back; keep the backpack, doll adornment, and clenched photograph clearly legible. Let soft dusk spill from the open window rake gently across the disorder while her figure remains naturally subdued against the doorway.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior with an open window.\n\nSHOT TEXT (authoritative, Korean): 백팩을 메고 구겨진 사진을 손에 꽉 쥔 채 열린 현관문 앞에 선 수리영의 뒷모습.\n\nLOCATION (lock): A compact rooftop apartment ransacked throughout, with footprints, overturned dining chairs, an open living-room window, and disturbed household items. The bedroom contains a bed and a wall marked by a red circle holding a bloodstained 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): Su-ri-yeong leaves wearing her doll-adorned backpack and tightly gripping the old photograph; the printed maps remain inside the pack and her phone remains with her. The ransacked room and red circle are left intact behind 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a close rear view from slightly below waist height, compressing the doorway around Su-ri-yeong and making the tension in her gripping hand the visual anchor. Keep her backpack and doll adornment prominent while the crumpled photograph is seen naturally from the camera-facing side dictated by her grip; allow only selective fragments of the disturbed apartment to remain behind her in shallow focus. Use low, cool dusk ambience with a soft edge from the open window, preserving tactile detail in her hand, the worn photograph, and the backpack.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S25sh7.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" }, { "label": "PROP REFERENCE — 인형이 달린 백팩: 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S25sh7.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" }, { "label": "PROP REFERENCE — 인형이 달린 백팩: 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" }, { "label": "PROP REFERENCE — 인형이 달린 백팩: 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" }, { "label": "PROP REFERENCE — 인형이 달린 백팩: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "C", "ranking": [ "C", "A", "B", "D" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "현관문 앞 뒷모습과 방을 등지고 떠나는 구도를 정확히 연출했으며, 소품(사진)의 기능적 면을 카메라에 억지로 노출하지 않아 규칙을 유일하게 준수함 (방 구조 변경은 3순위 감점)." }, { "label": "A", "score": 5, "verdict_ko": "방 구조는 참조와 완벽히 일치하나 현관문 앞이 아닌 주방을 향해 서 있으며, 사진 앞면을 카메라에 부자연스럽게 노출해 소품 방향 규칙을 위반함." }, { "label": "B", "score": 4, "verdict_ko": "방을 떠나는 설정과 반대로 실내를 향해 서 있고, 붉은 동그라미가 침실이 아닌 거실 벽에 위치하며 소품 방향 규칙도 위반함." }, { "label": "D", "score": 2, "verdict_ko": "거실과 침실 양쪽에 붉은 동그라미가 중복 생성되어 심각한 위반(Hard Violation)이며, 소품 방향 규칙과 떠나는 액션 구도 모두 실패함." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "A", "B", "D" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "현관문 앞 뒷모습과 방을 등지고 떠나는 구도를 정확히 연출했으며, 소품(사진)의 기능적 면을 카메라에 억지로 노출하지 않아 규칙을 유일하게 준수함 (방 구조 변경은 3순위 감점)." }, { "label": "A", "score": 5, "verdict_ko": "방 구조는 참조와 완벽히 일치하나 현관문 앞이 아닌 주방을 향해 서 있으며, 사진 앞면을 카메라에 부자연스럽게 노출해 소품 방향 규칙을 위반함." }, { "label": "B", "score": 4, "verdict_ko": "방을 떠나는 설정과 반대로 실내를 향해 서 있고, 붉은 동그라미가 침실이 아닌 거실 벽에 위치하며 소품 방향 규칙도 위반함." }, { "label": "D", "score": 2, "verdict_ko": "거실과 침실 양쪽에 붉은 동그라미가 중복 생성되어 심각한 위반(Hard Violation)이며, 소품 방향 규칙과 떠나는 액션 구도 모두 실패함." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "C", "B", "A" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "공간 구조와 지정된 소품들을 가장 정확하게 재현했으나, 사진을 꽉 쥔 손의 행동 연출이 다소 약합니다." }, { "label": "C", "score": 5, "verdict_ko": "문 앞에서의 행동 연출은 우수하나, 침실에 있어야 할 붉은 원이 거실 벽에 잘못 배치되는 공간 오류가 있습니다." }, { "label": "B", "score": 4, "verdict_ko": "기준 사진의 공간 구조를 완전히 무시하고 거실 안쪽 벽에 현관문을 배치한 심각한 오류가 있습니다." }, { "label": "A", "score": 3, "verdict_ko": "공간 구조가 전혀 다르며, 벽에 붉은 원 마커가 두 개나 중복 생성된 치명적인 오류가 있습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B", "C", "D" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "공간 구조와 지정된 소품들을 가장 정확하게 재현했으나, 사진을 꽉 쥔 손의 행동 연출이 다소 약합니다." }, { "label": "B", "score": 5, "verdict_ko": "문 앞에서의 행동 연출은 우수하나, 침실에 있어야 할 붉은 원이 거실 벽에 잘못 배치되는 공간 오류가 있습니다." }, { "label": "C", "score": 4, "verdict_ko": "기준 사진의 공간 구조를 완전히 무시하고 거실 안쪽 벽에 현관문을 배치한 심각한 오류가 있습니다." }, { "label": "D", "score": 3, "verdict_ko": "공간 구조가 전혀 다르며, 벽에 붉은 원 마커가 두 개나 중복 생성된 치명적인 오류가 있습니다." } ] }, "combined": { "totals": { "A": 12, "B": 9, "C": 12, "D": 5 }, "ranking": [ "A", "C", "B", "D" ], "agreement": false, "policy": 1 } }, "totals": { "A": 12, "B": 9, "C": 12, "D": 5 }, "selected": "A", "ranking": [ "A", "C", "B", "D" ], "verdicts": [ { "label": "C", "score": 8, "verdict_ko": "현관문 앞 뒷모습과 방을 등지고 떠나는 구도를 정확히 연출했으며, 소품(사진)의 기능적 면을 카메라에 억지로 노출하지 않아 규칙을 유일하게 준수함 (방 구조 변경은 3순위 감점)." }, { "label": "A", "score": 5, "verdict_ko": "방 구조는 참조와 완벽히 일치하나 현관문 앞이 아닌 주방을 향해 서 있으며, 사진 앞면을 카메라에 부자연스럽게 노출해 소품 방향 규칙을 위반함." }, { "label": "B", "score": 4, "verdict_ko": "방을 떠나는 설정과 반대로 실내를 향해 서 있고, 붉은 동그라미가 침실이 아닌 거실 벽에 위치하며 소품 방향 규칙도 위반함." }, { "label": "D", "score": 2, "verdict_ko": "거실과 침실 양쪽에 붉은 동그라미가 중복 생성되어 심각한 위반(Hard Violation)이며, 소품 방향 규칙과 떠나는 액션 구도 모두 실패함." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L05B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" }, { "label": "PROP REFERENCE — 인형이 달린 백팩: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "거실 창문이 열려 있지 않고 닫혀 있습니다.", "fix_en": "Slide the living room window open." }, { "issue_ko": "침실 벽의 붉은 원 안에 있어야 할 피 묻은 사진이 없습니다.", "fix_en": "Place the bloodstained photograph inside the red circle on the bedroom wall." }, { "issue_ko": "손에 쥔 사진의 앞면(이미지)이 수리영의 시선이 아닌 카메라를 향하고 있습니다.", "fix_en": "Flip the crumpled photograph in her hand so its blank back faces the camera." }, { "issue_ko": "수리영이 열린 현관문 앞이 아닌 방 안쪽 주방을 향해 서 있습니다.", "fix_en": "Rotate and reposition Su-ri-yeong to stand facing the open entrance door on the right." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Slide the living room window open.\n- Place the bloodstained photograph inside the red circle on the bedroom wall.\n- Flip the crumpled photograph in her hand so its blank back faces the camera.\n- Rotate and reposition Su-ri-yeong to stand facing the open entrance door on the right.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L05B01", "B": "L05B02", "C": "L05B03", "D": "L05B04", "E": "L05B05", "F": "L05B06", "G": "L05B07" }, "assigned": "L05B05", "choice": "Candidate E", "confident": true, "reason_ko": "\"열린 현관문 앞에 선\"이라는 지문에 따라 신발장과 출입문(현관)이 보이는 공간이 필요하므로 현재 할당된 컷이 적합합니다.", "kept": "L05B05" } }, "S26sh2::variants": { "author_fp": "5f95132fb1442e35", "author": { "variants": [ { "approach_ko": "낮은 후방 와이드 숏으로 수리영을 작게 두고 안개 낀 바다의 압도적인 여백과 고립감을 강조한다.", "prompt_en": "Photograph from a low, distant position directly behind her in a wide environmental composition. Place her seated figure low in the frame while the fog-obscured water occupies most of the image, using the tidal rocks, moorings, and landing point as subdued depth layers. Let the wet rock surfaces catch faint dusk reflections, with soft, directionless post-rain light preserving natural detail in her hair, clothing, body, and belongings rather than reducing her to a silhouette." }, { "approach_ko": "약간 높은 후방의 밀도 높은 중경 숏으로 굳은 자세와 젖은 갯바위의 질감을 가까이 포착한다.", "prompt_en": "Use a tighter rear view from slightly above her seated height, framing her head, shoulders, back, and grounded posture as the visual center while keeping the established belongings clearly readable within the composition. Compress the harbor and foggy water into quiet, overlapping planes behind her, with the wet tidal-rock texture held sharply in the foreground and the distance dissolving softly into mist. Shape the dusk illumination as a gentle lateral wash through the fog, emphasizing her absolute stillness without revealing her face." } ] }, "reused": false }, "S26sh2": { "input_fingerprint": "79bbb0cf395689d9", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 안개 낀 바다를 향해 갯바위 위에 멍하니 앉아 있는 수리영의 뒷모습.\n\nLOCATION (lock): An empty harbor bordered by tidal rocks, with moorings and a landing point for small boats. Dense fog obscures the water and hides other vessels from view. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong remains on the rocks with her backpack, the printed maps, the old photograph, and her mobile phone, which shows several missed calls from Hye-su.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 안개 낀 바다를 향해 갯바위 위에 멍하니 앉아 있는 수리영의 뒷모습.\n\nLOCATION (lock): An empty harbor bordered by tidal rocks, with moorings and a landing point for small boats. Dense fog obscures the water and hides other vessels from view. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong remains on the rocks with her backpack, the printed maps, the old photograph, and her mobile phone, which shows several missed calls from Hye-su.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph from a low, distant position directly behind her in a wide environmental composition. Place her seated figure low in the frame while the fog-obscured water occupies most of the image, using the tidal rocks, moorings, and landing point as subdued depth layers. Let the wet rock surfaces catch faint dusk reflections, with soft, directionless post-rain light preserving natural detail in her hair, clothing, body, and belongings 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 안개 낀 바다를 향해 갯바위 위에 멍하니 앉아 있는 수리영의 뒷모습.\n\nLOCATION (lock): An empty harbor bordered by tidal rocks, with moorings and a landing point for small boats. Dense fog obscures the water and hides other vessels from view. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong remains on the rocks with her backpack, the printed maps, the old photograph, and her mobile phone, which shows several missed calls from Hye-su.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a tighter rear view from slightly above her seated height, framing her head, shoulders, back, and grounded posture as the visual center while keeping the established belongings clearly readable within the composition. Compress the harbor and foggy water into quiet, overlapping planes behind her, with the wet tidal-rock texture held sharply in the foreground and the distance dissolving softly into mist. Shape the dusk illumination as a gentle lateral wash through the fog, emphasizing her absolute stillness without revealing her 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." }, "totals": { "A": 4, "B": 7 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "휴대폰 화면이 카메라를 향하게 들고 있어 소품 방향 규칙을 위반했으며, 오른손의 자세도 해부학적으로 부자연스럽습니다." }, { "label": "B", "score": 7, "verdict_ko": "뒷모습 구도와 바위 위의 소품 배치는 매우 훌륭하나, 안개가 옅어 배경에 배들이 노출된 점이 장소 설정과 어긋납니다." } ], "refs": [ { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프롬프트에 짙은 안개로 인해 다른 배들이 보이지 않는다고 명시되어 있으나, 우측 선착장과 좌측 배경 등에 여러 척의 배가 분명하게 나타나 있습니다.", "fix_en": "Remove all boats and vessels from the water and dock so that none are visible." }, { "issue_ko": "휴대폰 화면에 '부재중 전화' 내역이 표시되어야 하지만, 현재 전화가 걸려오고 있는 수신 화면(녹색/적색 버튼)이 떠 있습니다.", "fix_en": "Change the phone screen interface from an active incoming call to a screen displaying missed call text notifications." }, { "issue_ko": "캐릭터가 바다를 바라보고 앉아 있는데, 지도와 사진, 휴대폰 등 사용 중인 소지품들이 캐릭터의 시야 밖인 등 뒤(카메라 바로 앞)에 배치되어 있어 현실적인 방향성에 위배됩니다.", "fix_en": "Reposition the maps, photograph, and mobile phone to the rock space beside or slightly in front of the character so they fall within her natural field of view." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all boats and vessels from the water and dock so that none are visible.\n- Change the phone screen interface from an active incoming call to a screen displaying missed call text notifications.\n- Reposition the maps, photograph, and mobile phone to the rock space beside or slightly in front of the character so they fall within her natural field of view.\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", "share_plan": { "ref_plan": "background" } }, "S26sh6::variants": { "author_fp": "222ec5089aad7803", "author": { "variants": [ { "approach_ko": "수리영의 상체를 정면 중근경으로 고립시키고, 닻줄을 쥔 인우를 안개 낀 후경에 부드럽게 남기는 시선 높이 촬영.", "prompt_en": "Photograph the moment at eye level in an intimate medium close view centered on Su-ri-yeong’s upper body. Use shallow depth so her face, posture, and carried state command the frame while In-woo remains recognizable behind her with the mooring line in his hands, softened by the dense harbor fog. Let the cool, diffuse dusk light after rain model both figures gently, with restrained reflections from the wet surroundings." }, { "approach_ko": "인우 쪽에서 어깨 너머로 바라보며 닻줄을 쥔 손을 전경에 두고, 그 앞의 수리영 상체를 깊이감 있게 포착하는 구도.", "prompt_en": "Shoot from close beside In-woo in an over-the-shoulder composition, placing his line-holding hands and part of his figure in the immediate foreground while Su-ri-yeong’s upper body fills the opposing plane in front of him. Keep stronger depth definition so the physical relationship between her, In-woo, and the secured mooring line remains legible against the fog-muted harbor. Use the low-contrast dusk ambience and rain-darkened setting to create a tense, grounded sense of proximity." } ] }, "reused": false }, "S26sh6": { "input_fingerprint": "ff49278f8dc652c6", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 닻줄을 쥔 인우 앞에 선 수리영의 상체.\n\nLOCATION (lock): An empty harbor bordered by tidal rocks, with moorings and a landing point for small boats. Dense fog obscures the water and hides other vessels from view. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still carries her backpack, maps, old photograph, and phone as she approaches In-woo, who is holding and securing the mooring line.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 인우 (한국인 남성, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 닻줄을 쥔 인우 앞에 선 수리영의 상체.\n\nLOCATION (lock): An empty harbor bordered by tidal rocks, with moorings and a landing point for small boats. Dense fog obscures the water and hides other vessels from view. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still carries her backpack, maps, old photograph, and phone as she approaches In-woo, who is holding and securing the mooring line.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 인우 (한국인 남성, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment at eye level in an intimate medium close view centered on Su-ri-yeong’s upper body. Use shallow depth so her face, posture, and carried state command the frame while In-woo remains recognizable behind her with the mooring line in his hands, softened by the dense harbor fog. Let the cool, diffuse dusk light after rain model both figures gently, with restrained reflections from the wet 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.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 닻줄을 쥔 인우 앞에 선 수리영의 상체.\n\nLOCATION (lock): An empty harbor bordered by tidal rocks, with moorings and a landing point for small boats. Dense fog obscures the water and hides other vessels from view. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong still carries her backpack, maps, old photograph, and phone as she approaches In-woo, who is holding and securing the mooring line.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 인우 (한국인 남성, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nShoot from close beside In-woo in an over-the-shoulder composition, placing his line-holding hands and part of his figure in the immediate foreground while Su-ri-yeong’s upper body fills the opposing plane in front of him. Keep stronger depth definition so the physical relationship between her, In-woo, and the secured mooring line remains legible against the fog-muted harbor. Use the low-contrast dusk ambience and rain-darkened setting to create a tense, grounded sense of proximity.\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 샷 크기와 인물 배치를 완벽히 구현하고 모든 소품을 올바르게 쥐고 있으나, 인우의 장갑 재질이 레퍼런스와 다릅니다." }, { "label": "B", "score": 5, "verdict_ko": "인우의 장갑은 일치하지만, 휴대폰이 주머니가 아닌 겉옷 표면에 부자연스럽게 붙어 있는 물리적 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S26sh2_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": "인우가 입은 재킷의 오른쪽 가슴에 닻 로고가 나타나 있으나, 캐릭터 레퍼런스에서는 왼쪽 가슴에 있습니다.", "fix_en": "Remove the anchor logo from the right chest of In-woo's jacket to accurately match the character reference." }, { "issue_ko": "프롬프트의 지시와 달리, 수리영이 들고 있는 옛날 사진의 앞면(기능적인 면)이 뷰어를 향해 정면으로 노출되어 있습니다.", "fix_en": "Turn the old photograph so that its back faces the camera, hiding the picture side from the viewer." }, { "issue_ko": "인우가 밧줄을 쥐고 있는 위쪽 손(오른손)의 손가락 형태가 뭉개져 밧줄과 섞여 있습니다.", "fix_en": "Redraw In-woo's right gloved hand so the fingers wrap anatomically correctly around the mooring line." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the anchor logo from the right chest of In-woo's jacket to accurately match the character reference.\n- Turn the old photograph so that its back faces the camera, hiding the picture side from the viewer.\n- Redraw In-woo's right gloved hand so the fingers wrap anatomically correctly around the mooring line.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S26sh2" } }, "S26sh8::variants": { "author_fp": "1eef6738fca32b06", "author": { "variants": [ { "approach_ko": "눈높이의 밀착 측면 클로즈업으로 굳게 다문 입술과 통화의 긴장을 포착한다.", "prompt_en": "Photograph her in an intimate eye-level profile close-up, with the phone-side hand and the line of her firmly closed lips held in crisp focus. Use a gently compressed perspective and very shallow depth, letting the moored boat, wet harbor textures, and dense fog dissolve into muted dusk layers behind her. Keep the sea-facing side of the frame open so her fixed gaze and silence press into the negative space." }, { "approach_ko": "한발 물러선 측면 미디엄 숏으로 인물과 승선 직전의 항구 경계를 함께 긴장감 있게 구성한다.", "prompt_en": "Frame a restrained side-on medium shot from slightly below shoulder height, placing her beside the boat while preserving the landing point and fog-hidden water as readable spatial layers. Favor deeper focus and a strong foreground-to-background arrangement, using the boat edge and wet harbor geometry to hold her at the threshold before boarding. Let the soft, directional dusk glow trace her facial profile while the broad fog bank dominates the composition ahead of her." } ] }, "reused": false }, "S26sh8": { "input_fingerprint": "6b2c91ea4e38ac05", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 배에 오르기 전, 바다를 향해 귀에 핸드폰을 댄 채 입술을 굳게 다문 수리영의 옆얼굴.\n\nLOCATION (lock): An empty harbor bordered by tidal rocks, with moorings and a landing point for small boats. Dense fog obscures the water and hides other vessels from view. 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 dense harbor fog, moored boat, muted sunset light, and the young woman's continuing clothing and position beside the boat from the reference. Exclude the boatman and mooring rope from the tighter phone-call frame unless they remain naturally outside the edge.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Before boarding, Su-ri-yeong holds her phone to her ear to leave a voice message; her backpack, maps, and old photograph remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 배에 오르기 전, 바다를 향해 귀에 핸드폰을 댄 채 입술을 굳게 다문 수리영의 옆얼굴.\n\nLOCATION (lock): An empty harbor bordered by tidal rocks, with moorings and a landing point for small boats. Dense fog obscures the water and hides other vessels from view. 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 dense harbor fog, moored boat, muted sunset light, and the young woman's continuing clothing and position beside the boat from the reference. Exclude the boatman and mooring rope from the tighter phone-call frame unless they remain naturally outside the edge.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Before boarding, Su-ri-yeong holds her phone to her ear to leave a voice message; her backpack, maps, and old photograph remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph her in an intimate eye-level profile close-up, with the phone-side hand and the line of her firmly closed lips held in crisp focus. Use a gently compressed perspective and very shallow depth, letting the moored boat, wet harbor textures, and dense fog dissolve into muted dusk layers behind her. Keep the sea-facing side of the frame open so her fixed gaze and silence press into the negative space.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, dense fog after rain.\n\nSHOT TEXT (authoritative, Korean): 배에 오르기 전, 바다를 향해 귀에 핸드폰을 댄 채 입술을 굳게 다문 수리영의 옆얼굴.\n\nLOCATION (lock): An empty harbor bordered by tidal rocks, with moorings and a landing point for small boats. Dense fog obscures the water and hides other vessels from view. 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 dense harbor fog, moored boat, muted sunset light, and the young woman's continuing clothing and position beside the boat from the reference. Exclude the boatman and mooring rope from the tighter phone-call frame unless they remain naturally outside the edge.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Before boarding, Su-ri-yeong holds her phone to her ear to leave a voice message; her backpack, maps, and old photograph remain 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a restrained side-on medium shot from slightly below shoulder height, placing her beside the boat while preserving the landing point and fog-hidden water as readable spatial layers. Favor deeper focus and a strong foreground-to-background arrangement, using the boat edge and wet harbor geometry to hold her at the threshold before boarding. Let the soft, directional dusk glow trace her facial profile while the broad fog bank dominates the composition ahead of 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." }, "totals": { "A": 7, "B": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'옆얼굴'이라는 지시에 맞춰 클로즈업 프레임을 정확히 구현했으며 불필요한 배경 요소를 잘 배제했습니다." }, { "label": "B", "score": 4, "verdict_ko": "얼굴에 집중해야 하는 샷 구도를 무시하고 프레임을 넓혀 불필요한 하반신과 소품을 과도하게 노출했습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S26sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "핸드폰을 쥔 오른손의 손가락이 얼굴과 가까운 앞쪽 모서리에서 시작해 뒤쪽으로 감싸고 있어 해부학적으로 불가능한 자세입니다.", "fix_en": "Redraw the right hand so the knuckles are on the back edge of the phone with fingers wrapping forward towards the face." }, { "issue_ko": "핸드폰 후면 카메라 렌즈가 뒷머리 쪽 모서리에 있어 기기가 좌우 반전되어 있습니다 (실제 기기처럼 얼굴 쪽에 있어야 함).", "fix_en": "Move the rear camera module to the phone's top corner closest to the face to correct the mirrored orientation." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw the right hand so the knuckles are on the back edge of the phone with fingers wrapping forward towards the face.\n- Move the rear camera module to the phone's top corner closest to the face to correct the mirrored orientation.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S26sh6" } }, "S27sh1::variants": { "author_fp": "76e709a647767fb7", "author": { "variants": [ { "approach_ko": "눈높이의 밀도 높은 정면 미디엄 클로즈업으로 혜수의 얼어붙은 반응과 통화의 충격을 전면화한다.", "prompt_en": "Frame Hye-su in an eye-level, near-frontal medium close-up, with her upper body filling most of the image and the phone-hand profile clearly readable. Use a gently compressed perspective and selective focus: hold her face, ear, and phone in crisp detail while the ransacked room remains recognizable but softened behind her. Let the dusk window light fall laterally across her mature features, emphasizing absolute stillness, fixed gaze, and restrained facial tension." }, { "approach_ko": "비스듬한 측면의 환경적 상반신 숏으로 혜수를 화면 가장자리에 두고 난장판이 된 방의 깊이와 고립감을 강조한다.", "prompt_en": "Photograph the moment as an environmental upper-body shot from a slightly elevated three-quarter side angle, placing Hye-su off-center while the disordered room occupies the stronger share of the composition. Use a broader, layered perspective and deeper focus so the footprints, overturned chairs, open window, and red wall circle form a legible path through the frame around her rigid posture. Shape the dusk illumination from the window into restrained contrast, allowing the room's depth and emptiness to press visually against her frozen profile." } ] }, "reused": false }, "S27sh1": { "input_fingerprint": "f6811b6ae6651e1e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 엉망이 된 옥탑방 거실을 배경으로 핸드폰을 귀에 댄 채 굳어버린 혜수의 상체.\n\nLOCATION (lock): The same compact rooftop apartment interior, now empty enough for an investigator to stand and listen to a recorded phone message. 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 ransacked living-room state, sunset light, footprints, open window, and overturned dining chairs from the reference. Exclude the departing young woman, her backpack, and the photograph, replacing that moment with the investigator listening to the phone message.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 remains ransacked, with footprints, an open window, overturned chairs, and the red wall circle left after the photograph was removed. Hye-su holds her phone while listening to Su-ri-yeong’s message.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 엉망이 된 옥탑방 거실을 배경으로 핸드폰을 귀에 댄 채 굳어버린 혜수의 상체.\n\nLOCATION (lock): The same compact rooftop apartment interior, now empty enough for an investigator to stand and listen to a recorded phone message. 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 ransacked living-room state, sunset light, footprints, open window, and overturned dining chairs from the reference. Exclude the departing young woman, her backpack, and the photograph, replacing that moment with the investigator listening to the phone message.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 remains ransacked, with footprints, an open window, overturned chairs, and the red wall circle left after the photograph was removed. Hye-su holds her phone while listening to Su-ri-yeong’s message.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame Hye-su in an eye-level, near-frontal medium close-up, with her upper body filling most of the image and the phone-hand profile clearly readable. Use a gently compressed perspective and selective focus: hold her face, ear, and phone in crisp detail while the ransacked room remains recognizable but softened behind her. Let the dusk window light fall laterally across her mature features, emphasizing absolute stillness, fixed gaze, and restrained facial 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, interior lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 엉망이 된 옥탑방 거실을 배경으로 핸드폰을 귀에 댄 채 굳어버린 혜수의 상체.\n\nLOCATION (lock): The same compact rooftop apartment interior, now empty enough for an investigator to stand and listen to a recorded phone message. 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 ransacked living-room state, sunset light, footprints, open window, and overturned dining chairs from the reference. Exclude the departing young woman, her backpack, and the photograph, replacing that moment with the investigator listening to the phone message.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 remains ransacked, with footprints, an open window, overturned chairs, and the red wall circle left after the photograph was removed. Hye-su holds her phone while listening to Su-ri-yeong’s message.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 여성, 40대 중반, 성숙한 얼굴선, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment as an environmental upper-body shot from a slightly elevated three-quarter side angle, placing Hye-su off-center while the disordered room occupies the stronger share of the composition. Use a broader, layered perspective and deeper focus so the footprints, overturned chairs, open window, and red wall circle form a legible path through the frame around her rigid posture. Shape the dusk illumination from the window into restrained contrast, allowing the room's depth and emptiness to press visually against her frozen profile.\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 '상체' 샷의 크기를 정확히 구현했으며, 이전 샷의 구도를 복사하지 않고 인물의 굳은 연기에 올바르게 초점을 맞췄습니다." }, { "label": "B", "score": 4, "verdict_ko": "이전 레퍼런스의 카메라 구도를 그대로 복사하지 말라는 지침을 어겼으며, 요구된 상체 샷보다 너무 넓게 프레이밍되었습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S25sh7_sel.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "핸드폰의 화면(스크린)이 귀를 향하지 않고 카메라가 있는 바깥쪽을 향해 있습니다.", "fix_en": "Turn the cell phone so its screen is pressed against her ear and its back faces outward." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Turn the cell phone so its screen is pressed against her ear and its back 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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S25sh7" } }, "S28sh1::variants": { "author_fp": "39e5bbdb50f6a21f", "author": { "variants": [ { "approach_ko": "수평선과 붉게 물든 바다를 넓게 담고 어선을 작게 배치한 고요한 초광각 전경.", "prompt_en": "Photograph an extremely wide, restrained establishing view from near water level, with the open sea occupying most of the frame and the small fishing boat held modestly against the distant horizon. Use expansive depth and ample negative space to stress its isolation offshore, while the red dusk light skims across the water and gently outlines the exposed deck and enclosed wheelhouse." }, { "approach_ko": "높은 사선 시점과 망원 압축으로 어선의 갑판과 조타실을 선명하게 부각한 전경.", "prompt_en": "Use a high, oblique long-lens view that makes the fishing boat the dominant subject while compressing the surrounding open water into layered bands behind it. Frame the vessel diagonally so the exposed working deck leads toward the enclosed wheelhouse, with the low red dusk illumination shaping its surfaces and reflecting softly across the sea." } ] }, "reused": false }, "S28sh1": { "input_fingerprint": "0f6d40f5df272353", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 해질 무렵 붉은 노을빛이 비치는 바다 한가운데 소형 어선이 떠 있는 전경.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong and In-woo are aboard the same small fishing boat; her backpack, printed maps, old photograph, and phone remain aboard with her. The many circular marks already exist inside the wheelhouse.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 해질 무렵 붉은 노을빛이 비치는 바다 한가운데 소형 어선이 떠 있는 전경.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong and In-woo are aboard the same small fishing boat; her backpack, printed maps, old photograph, and phone remain aboard with her. The many circular marks already exist inside the wheelhouse.\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 an extremely wide, restrained establishing view from near water level, with the open sea occupying most of the frame and the small fishing boat held modestly against the distant horizon. Use expansive depth and ample negative space to stress its isolation offshore, while the red dusk light skims across the water and gently outlines the exposed deck and enclosed wheelhouse.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 해질 무렵 붉은 노을빛이 비치는 바다 한가운데 소형 어선이 떠 있는 전경.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong and In-woo are aboard the same small fishing boat; her backpack, printed maps, old photograph, and phone remain aboard with her. The many circular marks already exist inside the wheelhouse.\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, oblique long-lens view that makes the fishing boat the dominant subject while compressing the surrounding open water into layered bands behind it. Frame the vessel diagonally so the exposed working deck leads toward the enclosed wheelhouse, with the low red dusk illumination shaping its surfaces and reflecting softly across the sea.\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": "지문이 요구한 '바다 한가운데 소형 어선이 떠 있는 전경'을 알맞은 원경 샷으로 정확히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "어선의 전경을 보여주어야 하는 지문을 무시한 채 금지된 레퍼런스의 카메라 구도를 그대로 복사하여 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L20B01.png" } ], "critique": { "issues": [ { "issue_ko": "어선 뒤쪽 갑판 위에 사람들의 실루엣이 보입니다.", "fix_en": "Remove the silhouettes of the people standing on the back deck of the boat." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the silhouettes of the people standing on the back deck of the boat.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L20B01", "B": "L20B02" }, "assigned": "L20B02", "choice": "Candidate B", "confident": true, "reason_ko": "\"바다 한가운데 소형 어선이 떠 있는 전경\"을 묘사하고 있으며, 후보 이미지들이 동일한 어선 갑판 공간을 보여주므로 현재 할당된 후보를 유지합니다.", "kept": "L20B02" } }, "S28sh5::variants": { "author_fp": "453b947d0d0fcf70", "author": { "variants": [ { "approach_ko": "정면 아이레벨의 극단적 클로즈업으로 인우의 얼어붙은 당혹감과 크게 뜬 눈을 정적인 대칭 구도에 가둔다.", "prompt_en": "Photograph In-woo in an intense, eye-level frontal close-up, his face held nearly centered in a rigid, claustrophobic composition. Use a compressed lens feel and extremely shallow focus so his naturally widened eyes and frozen facial tension are razor-sharp while the boat setting falls into soft abstraction. Cool, diffuse dusk light models his features gently, preserving the stunned stillness of the beat." }, { "approach_ko": "얼굴 가까이 붙은 비스듬한 로우앵글과 넓은 원근감으로 갑작스러운 충격과 배 위의 불안정한 공간감을 강조한다.", "prompt_en": "Place the camera very close to In-woo at a slight low, off-axis angle, using a broader perspective that gives the close-up immediate, unsettled energy without distorting his anatomy. Frame his face asymmetrically, with a narrow trace of the fishing boat environment receding behind him to create tense depth. Let the remaining dusk light rake softly across one side of his face, emphasizing his arrested expression and fixed, startled gaze." } ] }, "reused": false }, "S28sh5": { "input_fingerprint": "fef54d2e0c23168e", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜨고 당황한 표정으로 굳어버린 인우의 얼굴 클로즈업.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old photograph remains in Su-ri-yeong’s hand after she shows it to In-woo, while her backpack and other maps remain aboard. The wheelhouse’s numerous circular marks are already present.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜨고 당황한 표정으로 굳어버린 인우의 얼굴 클로즈업.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old photograph remains in Su-ri-yeong’s hand after she shows it to In-woo, while her backpack and other maps remain aboard. The wheelhouse’s numerous circular marks are already present.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph In-woo in an intense, eye-level frontal close-up, his face held nearly centered in a rigid, claustrophobic composition. Use a compressed lens feel and extremely shallow focus so his naturally widened eyes and frozen facial tension are razor-sharp while the boat setting falls into soft abstraction. Cool, diffuse dusk light models his features gently, preserving the stunned stillness of the beat.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜨고 당황한 표정으로 굳어버린 인우의 얼굴 클로즈업.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old photograph remains in Su-ri-yeong’s hand after she shows it to In-woo, while her backpack and other maps remain aboard. The wheelhouse’s numerous circular marks are already present.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera very close to In-woo at a slight low, off-axis angle, using a broader perspective that gives the close-up immediate, unsettled energy without distorting his anatomy. Frame his face asymmetrically, with a narrow trace of the fishing boat environment receding behind him to create tense depth. Let the remaining dusk light rake softly across one side of his face, emphasizing his arrested expression and fixed, startled gaze.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜨고 당황한 표정으로 굳어버린 인우의 얼굴 클로즈업.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old photograph remains in Su-ri-yeong’s hand after she shows it to In-woo, while her backpack and other maps remain aboard. The wheelhouse’s numerous circular marks are already present.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph In-woo in an intense, eye-level frontal close-up, his face held nearly centered in a rigid, claustrophobic composition. Use a compressed lens feel and extremely shallow focus so his naturally widened eyes and frozen facial tension are razor-sharp while the boat setting falls into soft abstraction. Cool, diffuse dusk light models his features gently, preserving the stunned stillness of the beat.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 눈을 크게 뜨고 당황한 표정으로 굳어버린 인우의 얼굴 클로즈업.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old photograph remains in Su-ri-yeong’s hand after she shows it to In-woo, while her backpack and other maps remain aboard. The wheelhouse’s numerous circular marks are already present.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPlace the camera very close to In-woo at a slight low, off-axis angle, using a broader perspective that gives the close-up immediate, unsettled energy without distorting his anatomy. Frame his face asymmetrically, with a narrow trace of the fishing boat environment receding behind him to create tense depth. Let the remaining dusk light rake softly across one side of his face, emphasizing his arrested expression and fixed, startled gaze.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L20B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L20B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L20B01.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L20B01.png" }, { "label": "CHARACTER REFERENCE — 인우: 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": "D", "score": 7, "verdict_ko": "당황하여 굳은 얼굴 클로즈업을 잘 살렸으며, 조타실의 열린 문과 내부 조향장치 등 레퍼런스의 공간 디테일을 가장 정확히 재현했습니다." }, { "label": "C", "score": 6, "verdict_ko": "클로즈업 프레이밍과 표정 연기는 우수하나, 레퍼런스와 달리 조타실 문이 닫혀 있어 공간 재현도가 다소 떨어집니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 표정과 구도를 잘 따랐으나, 조타실 우측의 열린 문 구조가 생략되어 배경 일치도가 낮습니다." }, { "label": "B", "score": 3, "verdict_ko": "조타실 비율이 심하게 왜곡되고 구명환이 난간 위에 떠 있는 등 배경 구조에 물리적으로 불가능한 오류가 발생해 실격입니다." } ] }, "forward_normalized": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "당황하여 굳은 얼굴 클로즈업을 잘 살렸으며, 조타실의 열린 문과 내부 조향장치 등 레퍼런스의 공간 디테일을 가장 정확히 재현했습니다." }, { "label": "C", "score": 6, "verdict_ko": "클로즈업 프레이밍과 표정 연기는 우수하나, 레퍼런스와 달리 조타실 문이 닫혀 있어 공간 재현도가 다소 떨어집니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 표정과 구도를 잘 따랐으나, 조타실 우측의 열린 문 구조가 생략되어 배경 일치도가 낮습니다." }, { "label": "B", "score": 3, "verdict_ko": "조타실 비율이 심하게 왜곡되고 구명환이 난간 위에 떠 있는 등 배경 구조에 물리적으로 불가능한 오류가 발생해 실격입니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B", "D", "C" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지문이 요구한 표정과 클로즈업을 훌륭히 구현했으며, 배경 조타실의 열린 문과 구명환 위치가 레퍼런스와 완벽히 일치합니다." }, { "label": "B", "score": 5, "verdict_ko": "표정은 지문과 부합하나, 배경 조타실의 열린 문이 생략되는 등 장소 레퍼런스의 세부 구조가 부정확합니다." }, { "label": "C", "score": 3, "verdict_ko": "구명환이 우측 하단에 잘못 위치하는 등 배경 구조를 완전히 왜곡한 장소 오류가 심각합니다." }, { "label": "D", "score": 4, "verdict_ko": "인물의 앞머리 스타일이 레퍼런스와 다르고, 배경 선박의 형태와 비율이 원본 장소와 크게 다릅니다." } ] }, "reverse_normalized": { "winner": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "지문이 요구한 표정과 클로즈업을 훌륭히 구현했으며, 배경 조타실의 열린 문과 구명환 위치가 레퍼런스와 완벽히 일치합니다." }, { "label": "C", "score": 5, "verdict_ko": "표정은 지문과 부합하나, 배경 조타실의 열린 문이 생략되는 등 장소 레퍼런스의 세부 구조가 부정확합니다." }, { "label": "B", "score": 3, "verdict_ko": "구명환이 우측 하단에 잘못 위치하는 등 배경 구조를 완전히 왜곡한 장소 오류가 심각합니다." }, { "label": "A", "score": 4, "verdict_ko": "인물의 앞머리 스타일이 레퍼런스와 다르고, 배경 선박의 형태와 비율이 원본 장소와 크게 다릅니다." } ] }, "combined": { "totals": { "A": 9, "B": 6, "C": 11, "D": 14 }, "ranking": [ "D", "C", "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 9, "B": 6, "C": 11, "D": 14 }, "selected": "D", "ranking": [ "D", "C", "A", "B" ], "verdicts": [ { "label": "D", "score": 7, "verdict_ko": "당황하여 굳은 얼굴 클로즈업을 잘 살렸으며, 조타실의 열린 문과 내부 조향장치 등 레퍼런스의 공간 디테일을 가장 정확히 재현했습니다." }, { "label": "C", "score": 6, "verdict_ko": "클로즈업 프레이밍과 표정 연기는 우수하나, 레퍼런스와 달리 조타실 문이 닫혀 있어 공간 재현도가 다소 떨어집니다." }, { "label": "A", "score": 5, "verdict_ko": "지정된 표정과 구도를 잘 따랐으나, 조타실 우측의 열린 문 구조가 생략되어 배경 일치도가 낮습니다." }, { "label": "B", "score": 3, "verdict_ko": "조타실 비율이 심하게 왜곡되고 구명환이 난간 위에 떠 있는 등 배경 구조에 물리적으로 불가능한 오류가 발생해 실격입니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L20B01.png" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인우가 캐릭터 레퍼런스 이미지에서 착용하고 있는 남색 비니를 쓰지 않고 맨머리로 나타났습니다.", "fix_en": "Add the navy beanie from the character reference onto In-woo's head." }, { "issue_ko": "프롬프트에 해안선이 더 이상 가깝지 않은 먼 바다라고 명시되어 있으나, 배경 우측 수평선에 육지와 불빛이 뚜렷하게 보입니다.", "fix_en": "Remove the landmass and city lights from the right horizon to show only open water." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the navy beanie from the character reference onto In-woo's head.\n- Remove the landmass and city lights from the right horizon to show only open water.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L20B01", "B": "L20B02" }, "assigned": "L20B02", "choice": "Candidate B", "confident": true, "reason_ko": "인물의 얼굴 클로즈업 샷으로 특정 배경(갑판이나 조타실 등)이 명확히 요구되지 않으며, 두 후보가 동일한 공간을 보여주므로 현재 할당된 후보를 유지합니다.", "kept": "L20B02" } }, "S29sh4::variants": { "author_fp": "77b4097c2d1b9ca7", "author": { "variants": [ { "approach_ko": "수리영의 시선축에 가까운 정면 근접 촬영으로 사진 속 얼굴과 멈춘 손끝을 또렷하게 포착한다.", "prompt_en": "Photograph the hand and faded photograph in a tight, near-overhead close-up aligned close to Su-ri-yeong’s viewing axis, keeping the photograph’s front naturally facing her while clearly revealing the girl’s face beneath the arrested fingertip. Hold both fingertip texture and the worn photographic surface in crisp focus, with the wheelhouse falling into a restrained blur of dim dusk tones and soft circular forms." }, { "approach_ko": "사진 표면을 스치는 낮은 사선 각도와 얕은 초점으로 멈춰 선 손끝의 긴장감을 강조한다.", "prompt_en": "Use a low, oblique close-up skimming across the plane of the faded photograph, placing Su-ri-yeong’s motionless fingertip prominently in the near foreground and the girl’s face immediately beyond it. Let focus settle on the exact point of contact while the rest falls away quickly, with cool dusk window light grazing the hand and photograph and the wheelhouse’s circular shapes dissolving into distant highlights." } ] }, "reused": false }, "S29sh4": { "input_fingerprint": "d02fee4861c46cc6", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 빛바랜 사진 속 한 여자아이의 얼굴을 손가락 끝으로 짚은 채 멈춘 수리영의 손 클로즈업.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong holds the same faded photograph and points to the child identified as her mother. The wheelhouse walls remain covered with numerous circular marks, and her backpack stays 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 빛바랜 사진 속 한 여자아이의 얼굴을 손가락 끝으로 짚은 채 멈춘 수리영의 손 클로즈업.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong holds the same faded photograph and points to the child identified as her mother. The wheelhouse walls remain covered with numerous circular marks, and her backpack stays 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the hand and faded photograph in a tight, near-overhead close-up aligned close to Su-ri-yeong’s viewing axis, keeping the photograph’s front naturally facing her while clearly revealing the girl’s face beneath the arrested fingertip. Hold both fingertip texture and the worn photographic surface in crisp focus, with the wheelhouse falling into a restrained blur of dim dusk tones and soft circular forms.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 빛바랜 사진 속 한 여자아이의 얼굴을 손가락 끝으로 짚은 채 멈춘 수리영의 손 클로즈업.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong holds the same faded photograph and points to the child identified as her mother. The wheelhouse walls remain covered with numerous circular marks, and her backpack stays 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a low, oblique close-up skimming across the plane of the faded photograph, placing Su-ri-yeong’s motionless fingertip prominently in the near foreground and the girl’s face immediately beyond it. Let focus settle on the exact point of contact while the rest falls away quickly, with cool dusk window light grazing the hand and photograph and the wheelhouse’s circular shapes dissolving into distant highlights.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 빛바랜 사진 속 한 여자아이의 얼굴을 손가락 끝으로 짚은 채 멈춘 수리영의 손 클로즈업.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong holds the same faded photograph and points to the child identified as her mother. The wheelhouse walls remain covered with numerous circular marks, and her backpack stays 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the hand and faded photograph in a tight, near-overhead close-up aligned close to Su-ri-yeong’s viewing axis, keeping the photograph’s front naturally facing her while clearly revealing the girl’s face beneath the arrested fingertip. Hold both fingertip texture and the worn photographic surface in crisp focus, with the wheelhouse falling into a restrained blur of dim dusk tones and soft circular forms.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 빛바랜 사진 속 한 여자아이의 얼굴을 손가락 끝으로 짚은 채 멈춘 수리영의 손 클로즈업.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong holds the same faded photograph and points to the child identified as her mother. The wheelhouse walls remain covered with numerous circular marks, and her backpack stays 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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a low, oblique close-up skimming across the plane of the faded photograph, placing Su-ri-yeong’s motionless fingertip prominently in the near foreground and the girl’s face immediately beyond it. Let focus settle on the exact point of contact while the rest falls away quickly, with cool dusk window light grazing the hand and photograph and the wheelhouse’s circular shapes dissolving into distant highlights.\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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S29sh4.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S29sh4.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 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", "B", "D" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 낡은 사진 소품을 든 채 여자아이의 얼굴을 짚는 동작과 '손 클로즈업' 프레이밍을 가장 정확하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "소품과 동작은 일치하지만, 손 클로즈업이 아닌 인물의 얼굴과 어깨까지 포함된 넓은 프레이밍을 사용하여 감점되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍이 지시된 것보다 넓고 사진의 원근감이 어색하게 왜곡되었으며, 창문에 원형 자국이 덧씌워져 부자연스럽습니다." }, { "label": "D", "score": 3, "verdict_ko": "손 클로즈업 프레이밍은 우수하나, 레퍼런스와 완전히 다른 인물이 그려진 잘못된 사진 소품을 사용하여 치명적인 오류가 있습니다." } ] }, "forward_normalized": { "winner": "C", "ranking": [ "C", "A", "B", "D" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 낡은 사진 소품을 든 채 여자아이의 얼굴을 짚는 동작과 '손 클로즈업' 프레이밍을 가장 정확하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "소품과 동작은 일치하지만, 손 클로즈업이 아닌 인물의 얼굴과 어깨까지 포함된 넓은 프레이밍을 사용하여 감점되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍이 지시된 것보다 넓고 사진의 원근감이 어색하게 왜곡되었으며, 창문에 원형 자국이 덧씌워져 부자연스럽습니다." }, { "label": "D", "score": 3, "verdict_ko": "손 클로즈업 프레이밍은 우수하나, 레퍼런스와 완전히 다른 인물이 그려진 잘못된 사진 소품을 사용하여 치명적인 오류가 있습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "D", "C", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "지정된 사진 소품을 완전히 무시하고 다른 이미지를 생성했으며 손에 들고 있지도 않아 심각한 위반입니다." }, { "label": "B", "score": 8, "verdict_ko": "'손 클로즈업'이라는 샷 텍스트의 프레이밍을 가장 정확히 구현했으며, 지정된 소품과 장소를 훌륭하게 재현했습니다." }, { "label": "C", "score": 4, "verdict_ko": "가리키는 손가락이 사진과 융합되는 등 해부학적 오류가 있으며 사진 소품의 형태가 왜곡되었습니다." }, { "label": "D", "score": 6, "verdict_ko": "소품과 장소 재현은 우수하나, 손 클로즈업이 아닌 오버더숄더 샷으로 프레임을 넓혀 샷 텍스트 기준에서 감점되었습니다." } ] }, "reverse_normalized": { "winner": "C", "ranking": [ "C", "A", "B", "D" ], "verdicts": [ { "label": "D", "score": 3, "verdict_ko": "지정된 사진 소품을 완전히 무시하고 다른 이미지를 생성했으며 손에 들고 있지도 않아 심각한 위반입니다." }, { "label": "C", "score": 8, "verdict_ko": "'손 클로즈업'이라는 샷 텍스트의 프레이밍을 가장 정확히 구현했으며, 지정된 소품과 장소를 훌륭하게 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "가리키는 손가락이 사진과 융합되는 등 해부학적 오류가 있으며 사진 소품의 형태가 왜곡되었습니다." }, { "label": "A", "score": 6, "verdict_ko": "소품과 장소 재현은 우수하나, 손 클로즈업이 아닌 오버더숄더 샷으로 프레임을 넓혀 샷 텍스트 기준에서 감점되었습니다." } ] }, "combined": { "totals": { "A": 11, "B": 8, "C": 15, "D": 6 }, "ranking": [ "C", "A", "B", "D" ], "agreement": true, "policy": 1 } }, "totals": { "A": 11, "B": 8, "C": 15, "D": 6 }, "selected": "C", "ranking": [ "C", "A", "B", "D" ], "verdicts": [ { "label": "C", "score": 7, "verdict_ko": "지정된 낡은 사진 소품을 든 채 여자아이의 얼굴을 짚는 동작과 '손 클로즈업' 프레이밍을 가장 정확하게 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "소품과 동작은 일치하지만, 손 클로즈업이 아닌 인물의 얼굴과 어깨까지 포함된 넓은 프레이밍을 사용하여 감점되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍이 지시된 것보다 넓고 사진의 원근감이 어색하게 왜곡되었으며, 창문에 원형 자국이 덧씌워져 부자연스럽습니다." }, { "label": "D", "score": 3, "verdict_ko": "손 클로즈업 프레이밍은 우수하나, 레퍼런스와 완전히 다른 인물이 그려진 잘못된 사진 소품을 사용하여 치명적인 오류가 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 두 어린 여자아이와 시골 식당이 찍힌 빛바랜 낡은 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "수리영의 손가락 끝이 사진 속 여자아이의 얼굴이 아닌 식당 벽면을 짚고 있습니다.", "fix_en": "Move the pointing fingertip down so it rests directly on the face of the girl in the photograph." }, { "issue_ko": "사진의 앞면이 캐릭터의 시선이 아닌 카메라 렌즈 방향으로 부자연스럽게 틀어져 있습니다.", "fix_en": "Rotate the photograph so its front side naturally faces the eyes of the character holding it." }, { "issue_ko": "배경의 조타실 벽면에 명시된 수많은 원형 자국들이 보이지 않습니다.", "fix_en": "Add numerous circular marks to the visible wheelhouse walls in the blurred background." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the pointing fingertip down so it rests directly on the face of the girl in the photograph.\n- Rotate the photograph so its front side naturally faces the eyes of the character holding it.\n- Add numerous circular marks to the visible wheelhouse walls in the blurred background.\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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L21B01", "B": "L21B02" }, "assigned": "L21B01", "choice": "Candidate A", "confident": true, "reason_ko": "주어진 텍스트는 인물의 손과 사진에 집중하는 클로즈업 샷으로, 특정 서브공간을 구분할 만한 배경 요소가 명시되지 않았으므로 현재 할당된 후보를 유지합니다.", "kept": "L21B01" } }, "S29sh6::variants": { "author_fp": "d07efa31a33a44f1", "author": { "variants": [ { "approach_ko": "인우의 굳은 어깨와 뒤통수를 전경에 밀착시키고, 붉은 원형 표식으로 뒤덮인 벽면에 초점을 고정한 압박감 강한 오버숄더 쇼트.", "prompt_en": "Frame a tight eye-level over-the-shoulder view, with In-woo’s rigid shoulder and the back edge of his head looming close in the foreground while focus settles sharply on the wall crowded with red circular marks. Use compressed depth and restrained dusk ambience so the foreground feels immobilized and the marked wall presses toward him." }, { "approach_ko": "인우를 한쪽 전경에 작게 걸고 넓은 심도로 조타실 벽면의 붉은 원형 표식들이 공간을 잠식한 듯 보이게 하는 광각 시점.", "prompt_en": "Use a wider, slightly elevated over-the-shoulder composition that places In-woo’s tense shoulder and neatly groomed head along one edge while the marked wheelhouse walls dominate the frame. Hold broad depth of field across the cabin, arranging the dense red circles as a claustrophobic visual field around his fixed posture, with subdued directional dusk light shaping the interior." } ] }, "reused": false }, "S29sh6": { "input_fingerprint": "f507825c5c119307", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 굳어버린 인우의 어깨 너머, 조타실 벽면에 빼곡하게 그려진 기괴한 붉은색 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 numerous circular marks remain crowded across the wheelhouse walls, including the newly perceived red circle. Su-ri-yeong still possesses the old photograph, backpack, maps, and phone.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 굳어버린 인우의 어깨 너머, 조타실 벽면에 빼곡하게 그려진 기괴한 붉은색 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 numerous circular marks remain crowded across the wheelhouse walls, including the newly perceived red circle. Su-ri-yeong still possesses the old photograph, backpack, maps, and phone.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight eye-level over-the-shoulder view, with In-woo’s rigid shoulder and the back edge of his head looming close in the foreground while focus settles sharply on the wall crowded with red circular marks. Use compressed depth and restrained dusk ambience so the foreground feels immobilized and the marked wall presses toward him.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 굳어버린 인우의 어깨 너머, 조타실 벽면에 빼곡하게 그려진 기괴한 붉은색 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 numerous circular marks remain crowded across the wheelhouse walls, including the newly perceived red circle. Su-ri-yeong still possesses the old photograph, backpack, maps, and phone.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider, slightly elevated over-the-shoulder composition that places In-woo’s tense shoulder and neatly groomed head along one edge while the marked wheelhouse walls dominate the frame. Hold broad depth of field across the cabin, arranging the dense red circles as a claustrophobic visual field around his fixed posture, with subdued directional dusk light shaping the 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 굳어버린 인우의 어깨 너머, 조타실 벽면에 빼곡하게 그려진 기괴한 붉은색 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 numerous circular marks remain crowded across the wheelhouse walls, including the newly perceived red circle. Su-ri-yeong still possesses the old photograph, backpack, maps, and phone.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight eye-level over-the-shoulder view, with In-woo’s rigid shoulder and the back edge of his head looming close in the foreground while focus settles sharply on the wall crowded with red circular marks. Use compressed depth and restrained dusk ambience so the foreground feels immobilized and the marked wall presses toward him.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 굳어버린 인우의 어깨 너머, 조타실 벽면에 빼곡하게 그려진 기괴한 붉은색 원형 표식들이 보이는 시점 쇼트.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 numerous circular marks remain crowded across the wheelhouse walls, including the newly perceived red circle. Su-ri-yeong still possesses the old photograph, backpack, maps, and phone.\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, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a wider, slightly elevated over-the-shoulder composition that places In-woo’s tense shoulder and neatly groomed head along one edge while the marked wheelhouse walls dominate the frame. Hold broad depth of field across the cabin, arranging the dense red circles as a claustrophobic visual field around his fixed posture, with subdued directional dusk light shaping the 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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S29sh6.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/conti/conti_S29sh6.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.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", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'어깨 너머' 숄더 쇼트 구도와 벽면의 붉은 표식들을 정확히 연출했고, 인우의 복장(비니, 재킷)도 레퍼런스와 일치합니다." }, { "label": "B", "score": 3, "verdict_ko": "인우에게 설정에 없는 배낭을 추가하는 심각한 오류를 범했으며 비니도 누락되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "'어깨 너머'가 아닌 뒤통수를 정면에서 잡은 구도로 시야를 가리며, 비니가 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "지정된 오버 더 숄더 시점 쇼트를 무시한 측면 배치이며, 인우의 비니가 누락되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'어깨 너머' 숄더 쇼트 구도와 벽면의 붉은 표식들을 정확히 연출했고, 인우의 복장(비니, 재킷)도 레퍼런스와 일치합니다." }, { "label": "B", "score": 3, "verdict_ko": "인우에게 설정에 없는 배낭을 추가하는 심각한 오류를 범했으며 비니도 누락되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "'어깨 너머'가 아닌 뒤통수를 정면에서 잡은 구도로 시야를 가리며, 비니가 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "지정된 오버 더 숄더 시점 쇼트를 무시한 측면 배치이며, 인우의 비니가 누락되었습니다." } ] }, "reverse_raw": { "winner": "D", "ranking": [ "D", "B", "A", "C" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "어깨 너머 구도 지시를 어기고 측면 얼굴을 보여주었으며 캐릭터의 비니가 누락되었습니다." }, { "label": "B", "score": 5, "verdict_ko": "어깨 너머가 아닌 정후면 앵글로 연출되었고 캐릭터의 비니가 누락되었습니다." }, { "label": "C", "score": 3, "verdict_ko": "레퍼런스에 없는 배낭이 추가되었고, 붉은 표식이 벽이 아닌 창문에 잘못 배치되었습니다." }, { "label": "D", "score": 7, "verdict_ko": "정확한 어깨 너머(OTS) 구도를 구현했으며 레퍼런스의 의상(비니)을 유일하게 반영했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "D", "score": 4, "verdict_ko": "어깨 너머 구도 지시를 어기고 측면 얼굴을 보여주었으며 캐릭터의 비니가 누락되었습니다." }, { "label": "C", "score": 5, "verdict_ko": "어깨 너머가 아닌 정후면 앵글로 연출되었고 캐릭터의 비니가 누락되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스에 없는 배낭이 추가되었고, 붉은 표식이 벽이 아닌 창문에 잘못 배치되었습니다." }, { "label": "A", "score": 7, "verdict_ko": "정확한 어깨 너머(OTS) 구도를 구현했으며 레퍼런스의 의상(비니)을 유일하게 반영했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 6, "C": 9, "D": 8 }, "ranking": [ "A", "C", "D", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 6, "C": 9, "D": 8 }, "selected": "A", "ranking": [ "A", "C", "D", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "'어깨 너머' 숄더 쇼트 구도와 벽면의 붉은 표식들을 정확히 연출했고, 인우의 복장(비니, 재킷)도 레퍼런스와 일치합니다." }, { "label": "B", "score": 3, "verdict_ko": "인우에게 설정에 없는 배낭을 추가하는 심각한 오류를 범했으며 비니도 누락되었습니다." }, { "label": "C", "score": 4, "verdict_ko": "'어깨 너머'가 아닌 뒤통수를 정면에서 잡은 구도로 시야를 가리며, 비니가 누락되었습니다." }, { "label": "D", "score": 4, "verdict_ko": "지정된 오버 더 숄더 시점 쇼트를 무시한 측면 배치이며, 인우의 비니가 누락되었습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L21B02.png" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "정면 우측 창문에 부착된 원형 장치(선회창)가 엉뚱하게도 작은 조타 핸들 모양으로 잘못 생성되었습니다.", "fix_en": "Replace the small steering wheel attached to the front right window with a proper circular clear-view screen apparatus as seen in the location reference." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the small steering wheel attached to the front right window with a proper circular clear-view screen apparatus as seen in the location 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)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L21B01", "B": "L21B02" }, "assigned": "L21B01", "choice": "Candidate A", "confident": true, "reason_ko": "두 후보 이미지가 동일하여 조타실 벽면의 특정 요소를 구분할 명확한 차이가 없으므로 현재 할당된 후보를 유지함.", "kept": "L21B01" } }, "S29sh9::variants": { "author_fp": "e0994c9e6e088c15", "author": { "variants": [ { "approach_ko": "인우의 냉정한 눈과 어깨 너머로 처진 수리영의 옆얼굴을 한 평면에 압축한 밀착 클로즈업.", "prompt_en": "Photograph an intimate, eye-level close-up from slightly forward and to the side of In-woo, compressing his cold, fixed gaze and Su-ri-yeong’s drooping side profile over his shoulder into the same narrow plane. Crop tightly around their faces and his shoulder, with shallow focus holding both expressions legible while the cramped cabin and circular forms fall into a soft, sunset-tinted background. Let the window-side dusk light rake gently across their profiles, emphasizing his controlled tension against her complete slackness." }, { "approach_ko": "선실의 깊이와 원형 요소 사이에 두 사람을 겹쳐 놓고, 인우의 전방 응시와 수리영의 무력한 낙하를 대비시키는 측면 미디엄숏.", "prompt_en": "Use a restrained side-on medium shot from within the wheelhouse, framing In-woo in firm profile as he stares toward the bow while Su-ri-yeong’s limp head and side face hang visibly beyond his shoulder. Keep more of the steering area, windows, and repeated circular cabin details in view, arranging them in layered depth around the pair rather than isolating them. Favor deeper focus and directional dusk illumination from the windows, making his forward drive and her gravity-heavy collapse read through the opposing lines of their bodies." } ] }, "reused": false }, "S29sh9": { "input_fingerprint": "6fc1f0e6ddd3b170", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 차가운 시선으로 앞을 노려보는 인우의 어깨 위로 고개를 떨군 채 축 늘어진 수리영의 옆얼굴.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the cramped wheelhouse materials, sunset-tinted light, circular wall markings, and the man's continuing clothing and position from the reference. Exclude the pointing hand and old photograph, and frame the unconscious young woman collapsed over his shoulder.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Soo-ri-young is collapsed forward onto In-woo’s shoulder, with her torso supported against him, her head drooping beside his shoulder, and her limbs hanging limp toward the floor.\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): Su-ri-yeong is unconscious and is collapsed forward onto In-woo’s shoulder, with her torso supported against him, her head drooping beside his shoulder, and her limbs hanging limp toward the floor, unable to change her pose herself. Her backpack and personal items, including the old photograph, maps, and phone, remain aboard unless moved by someone else.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 인우 (한국인 남성, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 차가운 시선으로 앞을 노려보는 인우의 어깨 위로 고개를 떨군 채 축 늘어진 수리영의 옆얼굴.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the cramped wheelhouse materials, sunset-tinted light, circular wall markings, and the man's continuing clothing and position from the reference. Exclude the pointing hand and old photograph, and frame the unconscious young woman collapsed over his shoulder.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Soo-ri-young is collapsed forward onto In-woo’s shoulder, with her torso supported against him, her head drooping beside his shoulder, and her limbs hanging limp toward the floor.\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): Su-ri-yeong is unconscious and is collapsed forward onto In-woo’s shoulder, with her torso supported against him, her head drooping beside his shoulder, and her limbs hanging limp toward the floor, unable to change her pose herself. Her backpack and personal items, including the old photograph, maps, and phone, remain aboard unless moved by someone else.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 인우 (한국인 남성, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph an intimate, eye-level close-up from slightly forward and to the side of In-woo, compressing his cold, fixed gaze and Su-ri-yeong’s drooping side profile over his shoulder into the same narrow plane. Crop tightly around their faces and his shoulder, with shallow focus holding both expressions legible while the cramped cabin and circular forms fall into a soft, sunset-tinted background. Let the window-side dusk light rake gently across their profiles, emphasizing his controlled tension against her complete slackness.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, wheelhouse lighting unspecified.\n\nSHOT TEXT (authoritative, Korean): 차가운 시선으로 앞을 노려보는 인우의 어깨 위로 고개를 떨군 채 축 늘어진 수리영의 옆얼굴.\n\nLOCATION (lock): The enclosed wheelhouse of a small fishing boat, fitted with steering controls and windows overlooking the bow. Numerous circular shapes are visible throughout the cabin. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the cramped wheelhouse materials, sunset-tinted light, circular wall markings, and the man's continuing clothing and position from the reference. Exclude the pointing hand and old photograph, and frame the unconscious young woman collapsed over his shoulder.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Soo-ri-young is collapsed forward onto In-woo’s shoulder, with her torso supported against him, her head drooping beside his shoulder, and her limbs hanging limp toward the floor.\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): Su-ri-yeong is unconscious and is collapsed forward onto In-woo’s shoulder, with her torso supported against him, her head drooping beside his shoulder, and her limbs hanging limp toward the floor, unable to change her pose herself. Her backpack and personal items, including the old photograph, maps, and phone, remain aboard unless moved by someone else.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 인우 (한국인 남성, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained side-on medium shot from within the wheelhouse, framing In-woo in firm profile as he stares toward the bow while Su-ri-yeong’s limp head and side face hang visibly beyond his shoulder. Keep more of the steering area, windows, and repeated circular cabin details in view, arranging them in layered depth around the pair rather than isolating them. Favor deeper focus and directional dusk illumination from the windows, making his forward drive and her gravity-heavy collapse read through the opposing lines of their bodies.\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": "요구된 '어깨 위 수리영의 옆얼굴' 구도와 축 늘어진 자세를 훌륭히 구현했으나, 인우의 겉옷이 레퍼런스와 다르게 묘사된 점이 아쉽습니다." }, { "label": "B", "score": 3, "verdict_ko": "수리영의 머리가 어깨가 아닌 등 뒤에 비정상적으로 위치해 인체 해부학적 오류가 발생했으며, 제시된 구도를 충족하지 못했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S29sh6_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": "두 인물의 옷이 서로 바뀌었습니다 (인우가 회색 후드 재킷을, 수리영이 올리브 그린 재킷을 입고 있음).", "fix_en": "Swap their garments so In-woo wears his olive green jacket and Su-ri-yeong wears her dark grey hooded jacket." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Swap their garments so In-woo wears his olive green jacket and Su-ri-yeong wears her dark grey hooded 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.", "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S29sh6" } }, "S30sh3::variants": { "author_fp": "d131d9dd5891f23a", "author": { "variants": [ { "approach_ko": "수면 가까운 정면 대칭 구도로 안개에 잠기는 선수와 양옆 물보라의 압박감을 강조한다.", "prompt_en": "Photograph from very low near water level in a centered, head-on composition, with the hull rising forcefully above the camera and the bow’s entry into the fog as the dominant visual beat. Use a compressed lens feel so the dense mist forms an imposing wall behind the fixed white spray on both sides, while the golden reflection leads directly toward the vessel. Keep the light soft and backlit through the fog, preserving crisp spray against muted sea and mist." }, { "approach_ko": "멀리서 본 높은 사선 와이드 구도로 황금빛 항적과 거대한 안개층 속 배의 고립감을 부각한다.", "prompt_en": "Use a distant, elevated oblique wide view that places the vessel deeper in the frame and gives the surrounding sea and fog bank strong compositional weight. Arrange the long golden reflection as a diagonal path toward the boat, while the two white bursts beside the bow provide the sharpest visual accents as it crosses into the mist. Let atmospheric depth progressively soften the hull toward the fog, creating a restrained, ominous sense of isolation." } ] }, "reused": false }, "S30sh3": { "input_fingerprint": "d9f8fb56bf2947ee", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): sunset, golden reflection and dense fog.\n\nSHOT TEXT (authoritative, Korean): 뱃머리가 짙은 안개 속으로 반쯤 들어간 채 선수 양옆으로 흰 물보라가 고정되어 있는 소형 어선의 전경.\n\nLOCATION (lock): Open sea surrounding a small fishing boat as it approaches a dense bank of fog. The low sun casts a long golden reflection across the water before the vessel enters the mist. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 same boat continues into the dense fog with unconscious Su-ri-yeong aboard in In-woo’s custody. Her backpack, old photograph, maps, and phone remain on the vessel, and the wheelhouse circles remain in place.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): sunset, golden reflection and dense fog.\n\nSHOT TEXT (authoritative, Korean): 뱃머리가 짙은 안개 속으로 반쯤 들어간 채 선수 양옆으로 흰 물보라가 고정되어 있는 소형 어선의 전경.\n\nLOCATION (lock): Open sea surrounding a small fishing boat as it approaches a dense bank of fog. The low sun casts a long golden reflection across the water before the vessel enters the mist. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 same boat continues into the dense fog with unconscious Su-ri-yeong aboard in In-woo’s custody. Her backpack, old photograph, maps, and phone remain on the vessel, and the wheelhouse circles remain in place.\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 low near water level in a centered, head-on composition, with the hull rising forcefully above the camera and the bow’s entry into the fog as the dominant visual beat. Use a compressed lens feel so the dense mist forms an imposing wall behind the fixed white spray on both sides, while the golden reflection leads directly toward the vessel. Keep the light soft and backlit through the fog, preserving crisp spray against muted sea and mist.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): sunset, golden reflection and dense fog.\n\nSHOT TEXT (authoritative, Korean): 뱃머리가 짙은 안개 속으로 반쯤 들어간 채 선수 양옆으로 흰 물보라가 고정되어 있는 소형 어선의 전경.\n\nLOCATION (lock): Open sea surrounding a small fishing boat as it approaches a dense bank of fog. The low sun casts a long golden reflection across the water before the vessel enters the mist. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\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 same boat continues into the dense fog with unconscious Su-ri-yeong aboard in In-woo’s custody. Her backpack, old photograph, maps, and phone remain on the vessel, and the wheelhouse circles remain in place.\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, elevated oblique wide view that places the vessel deeper in the frame and gives the surrounding sea and fog bank strong compositional weight. Arrange the long golden reflection as a diagonal path toward the boat, while the two white bursts beside the bow provide the sharpest visual accents as it crosses into the mist. Let atmospheric depth progressively soften the hull toward the fog, creating a restrained, ominous sense of isolation.\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": 4 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "뱃머리가 짙은 안개 속으로 반쯤 들어간 전경과 황금빛 윤슬, 선수 양옆의 물보라를 텍스트에 맞게 훌륭히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "안개 속으로 '반쯤 들어간' 연출이 부족하며, 어선의 외관이 레퍼런스의 재질(흰색 철선)과 다르게 묘사되었습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/episodes/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/images/background_chain/L19B01.png" } ], "critique": { "issues": [ { "issue_ko": "뱃머리가 짙은 안개 속으로 반쯤 들어가 있어야 하나, 어선 전체가 안개 밖에 완전히 노출되어 있습니다.", "fix_en": "Move the boat forward so its front half is visibly engulfed by the dense fog." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move the boat forward so its front half is visibly engulfed by the dense fog.\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)", "share_plan": { "ref_plan": "background" } }, "S7sh5::variants": { "author_fp": "8eb55ecb9067c05d", "author": { "variants": [ { "approach_ko": "눈높이의 타이트한 측면 클로즈업으로 얼굴의 긴장과 통화에 집중하고 뒤쪽 철장은 얕은 심도로 흐린다.", "prompt_en": "Photograph this as an intimate eye-level side-profile close-up, cropping tightly around her head and upper shoulders. Keep her furrowed brow, cheek, and the phone at her ear in crisp focus while the barred enclosures and dogs behind her fall into soft, recognizable depth; use restrained available daytime light and subtle facial modeling to hold the tension." }, { "approach_ko": "약간 낮은 중거리 환경 숏으로 수리영과 철장 속 개들을 깊은 초점의 층으로 함께 포착한다.", "prompt_en": "Use a slightly low, medium-wide lateral view that preserves her clear profile while giving the shelter environment substantial visual weight. Arrange her against the repeating enclosure geometry, with the dogs visibly occupying the depth behind her back; favor deeper focus, firm structural lines, and unembellished daytime illumination for a tense observational feeling." } ] }, "reused": false }, "S7sh5": { "input_fingerprint": "6e9cc9ec668848ac", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철장 안의 개들을 등지고 귀에 휴대폰을 댄 채 미간을 찌푸린 수리영의 측면\n\nLOCATION (lock): A dog shelter with secure enclosures for multiple dogs and a work area for vaccinations, grooming, ointment application, and bathing. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong has her mobile phone with her and holds it to her ear while trying to contact Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 로트와일러 (넓고 각진 머리, 짧고 굵은 주둥이, 검은 털, 황갈색 얼굴 무늬) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철장 안의 개들을 등지고 귀에 휴대폰을 댄 채 미간을 찌푸린 수리영의 측면\n\nLOCATION (lock): A dog shelter with secure enclosures for multiple dogs and a work area for vaccinations, grooming, ointment application, and bathing. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong has her mobile phone with her and holds it to her ear while trying to contact Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 로트와일러 (넓고 각진 머리, 짧고 굵은 주둥이, 검은 털, 황갈색 얼굴 무늬) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 an intimate eye-level side-profile close-up, cropping tightly around her head and upper shoulders. Keep her furrowed brow, cheek, and the phone at her ear in crisp focus while the barred enclosures and dogs behind her fall into soft, recognizable depth; use restrained available daytime light and subtle facial modeling to hold 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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철장 안의 개들을 등지고 귀에 휴대폰을 댄 채 미간을 찌푸린 수리영의 측면\n\nLOCATION (lock): A dog shelter with secure enclosures for multiple dogs and a work area for vaccinations, grooming, ointment application, and bathing. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong has her mobile phone with her and holds it to her ear while trying to contact Minsuk.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 로트와일러 (넓고 각진 머리, 짧고 굵은 주둥이, 검은 털, 황갈색 얼굴 무늬) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 slightly low, medium-wide lateral view that preserves her clear profile while giving the shelter environment substantial visual weight. Arrange her against the repeating enclosure geometry, with the dogs visibly occupying the depth behind her back; favor deeper focus, firm structural lines, and unembellished daytime illumination for a tense observational 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." }, "totals": { "B": 7, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요청된 측면(profile) 구도를 정확히 구현했으며, 휴대폰의 방향과 인물의 표정 등 모든 지시사항을 잘 따랐습니다." }, { "label": "A", "score": 3, "verdict_ko": "휴대폰의 화면이 바깥을 향하게 쥐고 있어 소품 사용 규칙을 위반했으며, 완전한 측면보다는 사선 구도에 가깝습니다." } ], "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/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/scene/recipe/S7sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "우측 하단 테이블 위의 흰색 상자에 텍스트가 있습니다.", "fix_en": "Remove all text from the white boxes on the table." } ] }, "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 white boxes on the table.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S7sh2" }, "lane_policy": "ab_select_ready:share_plan_prev_bypass" }, "S28sh9::variants": { "author_fp": "1148bba533a67cff", "author": { "variants": [ { "approach_ko": "갑판과 수평선을 넓게 담는 낮은 측면 와이드숏으로, 수리영의 기울어진 전신 윤곽과 뻗은 팔의 절박함을 강조한다.", "prompt_en": "Photograph the moment in a low, wide lateral view across the working deck, keeping Su-ri-yeong’s entire body clearly readable from head to feet. Use the dusk sky as soft backlight so her human form, clothing, hair, outstretched arm, and extreme forward lean register in a dark but naturally detailed silhouette; arrange the depth so the direction of her reach toward In-woo’s back is unmistakable. Hold enough surrounding deck and open water in frame to make her precarious posture feel exposed and urgent." }, { "approach_ko": "인우의 등 가까이에서 뒤돌아보는 압축된 역앵글로, 전경의 인우와 그 너머 수리영의 전신 실루엣 사이 긴장을 만든다.", "prompt_en": "Frame from close behind In-woo, using his back as a dominant near-foreground anchor while keeping Su-ri-yeong’s full figure visible farther along the deck. Compress the depth so her urgently extended arm seems to reach across the narrow separation toward him, with her strongly leaning body outlined by restrained dusk backlight. Keep the focus and tonal emphasis on her complete human posture while In-woo’s nearer back creates immediate physical tension and a sense of movement toward the wheelhouse." } ] }, "reused": false }, "S28sh9": { "input_fingerprint": "777dd2715da4e7f3", "prompt": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 인우의 등 뒤를 향해 다급하게 팔을 뻗은 채 몸을 크게 기울인 수리영의 전신 실루엣.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong follows In-woo toward the wheelhouse with the old photograph still in her possession; her backpack, maps, and phone remain aboard. The wheelhouse walls are already covered with circular marks.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 인우 (한국인 남성, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 인우의 등 뒤를 향해 다급하게 팔을 뻗은 채 몸을 크게 기울인 수리영의 전신 실루엣.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong follows In-woo toward the wheelhouse with the old photograph still in her possession; her backpack, maps, and phone remain aboard. The wheelhouse walls are already covered with circular marks.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 인우 (한국인 남성, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the moment in a low, wide lateral view across the working deck, keeping Su-ri-yeong’s entire body clearly readable from head to feet. Use the dusk sky as soft backlight so her human form, clothing, hair, outstretched arm, and extreme forward lean register in a dark but naturally detailed silhouette; arrange the depth so the direction of her reach toward In-woo’s back is unmistakable. Hold enough surrounding deck and open water in frame to make her precarious posture feel exposed and urgent.\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 — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 인우의 등 뒤를 향해 다급하게 팔을 뻗은 채 몸을 크게 기울인 수리영의 전신 실루엣.\n\nLOCATION (lock): A small fishing boat underway on open water, with an exposed working deck and an enclosed wheelhouse. The coastline is no longer nearby, suggesting the vessel has traveled well offshore. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-yeong follows In-woo toward the wheelhouse with the old photograph still in her possession; her backpack, maps, and phone remain aboard. The wheelhouse walls are already covered with circular marks.\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, 20대 초반, 앳된 타원형 얼굴, 자연스러운 검은 머리); 인우 (한국인 남성, 20대 중반, 젊고 단정한 얼굴, 자연스러운 검은 머리) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame from close behind In-woo, using his back as a dominant near-foreground anchor while keeping Su-ri-yeong’s full figure visible farther along the deck. Compress the depth so her urgently extended arm seems to reach across the narrow separation toward him, with her strongly leaning body outlined by restrained dusk backlight. Keep the focus and tonal emphasis on her complete human posture while In-woo’s nearer back creates immediate physical tension and a sense of movement toward the wheelhouse.\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": "낡은 사진을 손에 쥐고 인우의 등 뒤로 팔을 뻗은 수리영의 전신 실루엣과 조타실 벽의 원형 자국을 지문에 맞게 잘 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "수리영의 전신이 완전히 담기지 않았으며, 쥐고 있어야 할 낡은 사진이 없고 조타실 벽의 원형 자국도 누락되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/358f9d88-79ad-4cec-b320-dd5565e68fd6/images/a38f2479-d6d8-4f31-b4bd-5e2136d3e501/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": "수리영이 등에 백팩을 메고 있으나, 프롬프트의 'Carried State' 지시사항에 따르면 배낭은 그녀의 몸이 아닌 배 안에 따로 남아 있어야 합니다.", "fix_en": "Remove the backpack from Su-ri-yeong's back." } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the backpack from Su-ri-yeong's back.\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", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S28sh5" }, "lane_policy": "ab_select_ready:share_plan_prev_bypass" } }