{ "groupbg::산속 숲 현장": { "input_fingerprint": "58106e80bdbcbd73", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "산속 숲 현장", "tags": [ "S1sh3", "S3sh1" ] }, "context_sig": "324c3616688077f9" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: A dense patch of undergrowth within a mountain forest, at the spot where the fleeing victim is attacked outdoors at dusk.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월산 숲속 범죄 현장: 우거진 수풀과 흙바닥으로 이루어진 산속 현장으로, 낮에는 경찰 통제선이 쳐진 감식 구역으로 보인다. (특징: 무성한 나무와 수풀; 경찰 통제선(폴리스 라인); 흙바닥과 나뭇잎; 핏자국)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#1. 숲속 / 해 질 무렵\n- 씬1, 2 현장을 감식하는 경찰, 폴리스 라인을 치고 시신을 수거한다.\n\nTIME OF DAY (lock): dusk.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: A dense patch of undergrowth within a mountain forest, at the spot where the fleeing victim is attacked outdoors at dusk.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n금월산 숲속 범죄 현장: 우거진 수풀과 흙바닥으로 이루어진 산속 현장으로, 낮에는 경찰 통제선이 쳐진 감식 구역으로 보인다. (특징: 무성한 나무와 수풀; 경찰 통제선(폴리스 라인); 흙바닥과 나뭇잎; 핏자국)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#1. 숲속 / 해 질 무렵\n- 씬1, 2 현장을 감식하는 경찰, 폴리스 라인을 치고 시신을 수거한다.\n\nTIME OF DAY (lock): dusk.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_산속_숲_현장_3a6a3c.png", "asset_id": "8ad24788-9939-45c0-a31f-5af8b4948512", "input_asset_ids": [ "7060ab7e-cb9a-4f6d-a66f-33c0e6037194" ], "origin_tag": "S1sh3", "place_text": "A dense patch of undergrowth within a mountain forest, at the spot where the fleeing victim is attacked outdoors at dusk.", "origin_inputs": { "place_text": "A dense patch of undergrowth within a mountain forest, at the spot where the fleeing victim is attacked outdoors at dusk.", "time_of_day_en": "dusk", "conti_asset_id": "7060ab7e-cb9a-4f6d-a66f-33c0e6037194" } }, "S1sh3::bgfirst_bg": { "input_fingerprint": "057169f3eceaeeeb", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 허공에서 날아든 거무스름한 인영이 금월산 피해 남성의 어깨를 강하게 내리눌러 남자의 몸이 뒤로 크게 꺾이며 무너지는 충돌의 찰나.\n\nLOCATION (lock): A dense patch of undergrowth within a mountain forest, at the spot where the fleeing victim is attacked outdoors at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From upper-body height several steps away, the camera finishes its track on a three-quarter side view, tightening only around the shoulder impact while retaining the victim's torso and backward-breaking line. The dark figure enters across the left midground into the contact point near center, while the victim folds toward the right side of frame against the surrounding undergrowth.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 거무스름한 인영 in the middle-left of the frame, midground, moves toward fleeing victim's shoulder; 금월산 피해 남성 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (우거진 상태); used as Surrounds the collision and preserves the forest setting behind both figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key dusk light appropriate to the forest preserves the figures' separation while keeping the impact tense and tactile.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 허공에서 날아든 거무스름한 인영이 금월산 피해 남성의 어깨를 강하게 내리눌러 남자의 몸이 뒤로 크게 꺾이며 무너지는 충돌의 찰나.\n\nLOCATION (lock): A dense patch of undergrowth within a mountain forest, at the spot where the fleeing victim is attacked outdoors at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From upper-body height several steps away, the camera finishes its track on a three-quarter side view, tightening only around the shoulder impact while retaining the victim's torso and backward-breaking line. The dark figure enters across the left midground into the contact point near center, while the victim folds toward the right side of frame against the surrounding undergrowth.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 거무스름한 인영 in the middle-left of the frame, midground, moves toward fleeing victim's shoulder; 금월산 피해 남성 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (우거진 상태); used as Surrounds the collision and preserves the forest setting behind both figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key dusk light appropriate to the forest preserves the figures' separation while keeping the impact tense and tactile.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S1sh3__bgfirst_bg.png", "asset_id": "22cdf047-8894-4dca-a2a5-41766c746cfc", "input_asset_ids": [ "7060ab7e-cb9a-4f6d-a66f-33c0e6037194", "8ad24788-9939-45c0-a31f-5af8b4948512" ] }, "S1sh3": { "input_fingerprint": "cf723d4fb6c1b34e", "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): 허공에서 날아든 거무스름한 인영이 금월산 피해 남성의 어깨를 강하게 내리눌러 남자의 몸이 뒤로 크게 꺾이며 무너지는 충돌의 찰나.\n\nLOCATION (lock): A dense patch of undergrowth within a mountain forest, at the spot where the fleeing victim is attacked outdoors at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From upper-body height several steps away, the camera finishes its track on a three-quarter side view, tightening only around the shoulder impact while retaining the victim's torso and backward-breaking line. The dark figure enters across the left midground into the contact point near center, while the victim folds toward the right side of frame against the surrounding undergrowth.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 거무스름한 인영 in the middle-left of the frame, midground, moves toward fleeing victim's shoulder; 금월산 피해 남성 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (우거진 상태); used as Surrounds the collision and preserves the forest setting behind both figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key dusk light appropriate to the forest preserves the figures' separation while keeping the impact tense and tactile.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The mountain victim is still missing one shoe, with the large bite wound on his thigh continuing to bleed as the dark figure knocks him down.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 금월산 피해 남성 (한국인 성인 남성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 허공에서 날아든 거무스름한 인영이 금월산 피해 남성의 어깨를 강하게 내리눌러 남자의 몸이 뒤로 크게 꺾이며 무너지는 충돌의 찰나.\n\nLOCATION (lock): A dense patch of undergrowth within a mountain forest, at the spot where the fleeing victim is attacked outdoors at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From upper-body height several steps away, the camera finishes its track on a three-quarter side view, tightening only around the shoulder impact while retaining the victim's torso and backward-breaking line. The dark figure enters across the left midground into the contact point near center, while the victim folds toward the right side of frame against the surrounding undergrowth.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 거무스름한 인영 in the middle-left of the frame, midground, moves toward fleeing victim's shoulder; 금월산 피해 남성 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (우거진 상태); used as Surrounds the collision and preserves the forest setting behind both figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key dusk light appropriate to the forest preserves the figures' separation while keeping the impact tense and tactile.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The mountain victim is still missing one shoe, with the large bite wound on his thigh continuing to bleed as the dark figure knocks him down.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 금월산 피해 남성 (한국인 성인 남성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 허공에서 날아든 거무스름한 인영이 금월산 피해 남성의 어깨를 강하게 내리눌러 남자의 몸이 뒤로 크게 꺾이며 무너지는 충돌의 찰나.\n\nLOCATION (lock): A dense patch of undergrowth within a mountain forest, at the spot where the fleeing victim is attacked outdoors at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From upper-body height several steps away, the camera finishes its track on a three-quarter side view, tightening only around the shoulder impact while retaining the victim's torso and backward-breaking line. The dark figure enters across the left midground into the contact point near center, while the victim folds toward the right side of frame against the surrounding undergrowth.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 거무스름한 인영 in the middle-left of the frame, midground, moves toward fleeing victim's shoulder; 금월산 피해 남성 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 우거진 수풀 (우거진 상태); used as Surrounds the collision and preserves the forest setting behind both figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key dusk light appropriate to the forest preserves the figures' separation while keeping the impact tense and tactile.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The mountain victim is still missing one shoe, with the large bite wound on his thigh continuing to bleed as the dark figure knocks him down.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 금월산 피해 남성 (한국인 성인 남성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S1sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S1sh3.png" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_산속_숲_현장_3a6a3c.png" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 '신발 한 짝을 잃어버린' 상태와 '거무스름한 인영'의 묘사를 비교적 정확히 반영하여 프롬프트 충실도가 높습니다." }, { "label": "A", "score": 4, "verdict_ko": "피해자가 두 신발을 모두 벗고 있으며, 공격자가 '거무스름한 인영'이라는 묘사와 달리 너무 밝게 드러나 감점되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 '신발 한 짝을 잃어버린' 상태와 '거무스름한 인영'의 묘사를 비교적 정확히 반영하여 프롬프트 충실도가 높습니다." }, { "label": "A", "score": 4, "verdict_ko": "피해자가 두 신발을 모두 벗고 있으며, 공격자가 '거무스름한 인영'이라는 묘사와 달리 너무 밝게 드러나 감점되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "지정된 미디엄 샷(Medium shot)이 아닌 전신이 보이는 풀 샷으로 연출되었으나, 공격자가 '거무스름한 인영'으로 묘사되었고 신발 착용 상태가 비교적 지시와 부합합니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍 지시(미디엄 샷)를 어긴 넓은 샷이며, 피해자가 신발을 모두 벗고 있어 '신발 한 짝을 잃어버린' 상태 지시를 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 5, "verdict_ko": "지정된 미디엄 샷(Medium shot)이 아닌 전신이 보이는 풀 샷으로 연출되었으나, 공격자가 '거무스름한 인영'으로 묘사되었고 신발 착용 상태가 비교적 지시와 부합합니다." }, { "label": "A", "score": 4, "verdict_ko": "프레이밍 지시(미디엄 샷)를 어긴 넓은 샷이며, 피해자가 신발을 모두 벗고 있어 '신발 한 짝을 잃어버린' 상태 지시를 위반했습니다." } ] }, "combined": { "totals": { "A": 8, "B": 12 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 8, "B": 12 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 '신발 한 짝을 잃어버린' 상태와 '거무스름한 인영'의 묘사를 비교적 정확히 반영하여 프롬프트 충실도가 높습니다." }, { "label": "A", "score": 4, "verdict_ko": "피해자가 두 신발을 모두 벗고 있으며, 공격자가 '거무스름한 인영'이라는 묘사와 달리 너무 밝게 드러나 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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_산속_숲_현장_3a6a3c.png" } ], "critique": { "issues": [ { "issue_ko": "카메라 구도가 지정된 미디엄 샷이 아니라 인물들의 전신이 모두 보이는 풀 샷으로 설정되었습니다.", "fix_en": "Move the camera closer to create a medium shot that focuses on the torsos and the shoulder impact.", "unfixable": true }, { "issue_ko": "거무스름한 인영이 허공에서 날아드는 모습이 아니라 두 발을 땅에 딛고 서 있습니다.", "fix_en": "Redraw the dark figure so they are suspended in mid-air, leaping downwards toward the victim.", "unfixable": false }, { "issue_ko": "거무스름한 인영이 피해자의 어깨를 강하게 내리누르지 않고 목을 움켜쥐고 있습니다.", "fix_en": "Reposition the dark figure's right hand to press heavily down on the victim's shoulder instead of grabbing his neck.", "unfixable": false }, { "issue_ko": "피해 남성이 신발 한 짝을 잃어버린 상태여야 하나, 양쪽 발에 모두 신발을 신고 있습니다.", "fix_en": "Remove the shoe from the victim's forward foot, replacing it with a bare foot or sock.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Crop in to a medium shot centered on the shoulder impact, retaining the victim’s torso and backward-breaking body line while reducing the surrounding empty forest area.", "issue_ko": "프레임이 두 인물의 전신과 넓은 숲을 담아 지정된 미디엄 숏보다 지나치게 넓다.", "unfixable": false }, { "fix_en": "Shift the dark figure’s striking hand from the victim’s neck to the top of his shoulder while preserving both figures’ positions and motion axes.", "issue_ko": "충돌 지점이 피해 남성의 어깨가 아니라 목 부근에 놓여 있다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw the dark figure so they are suspended in mid-air, leaping downwards toward the victim.\n- Reposition the dark figure's right hand to press heavily down on the victim's shoulder instead of grabbing his neck.\n- Remove the shoe from the victim's forward foot, replacing it with a bare foot or sock.\n- Crop in to a medium shot centered on the shoulder impact, retaining the victim’s torso and backward-breaking body line while reducing the surrounding empty forest area.\n- Shift the dark figure’s striking hand from the victim’s neck to the top of his shoulder while preserving both figures’ positions and motion axes.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "요구된 미디엄 샷 프레이밍과 공중에서 날아드는 동작은 충족하지 못했으나, 허벅지의 교상 출혈을 잘 반영했으며 치명적인 구조적 오류가 없어 B보다 우세합니다." }, { "label": "B", "score": 3, "verdict_ko": "공중에서 날아드는 동작은 표현되었으나, 가해자 등 쪽에 정체불명의 기계/금속 물체가 생성된 하드 위반(invented object)이 있으며 지정된 허벅지 상처도 누락되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "요구된 미디엄 샷 프레이밍과 공중에서 날아드는 동작은 충족하지 못했으나, 허벅지의 교상 출혈을 잘 반영했으며 치명적인 구조적 오류가 없어 B보다 우세합니다." }, { "label": "B", "score": 3, "verdict_ko": "공중에서 날아드는 동작은 표현되었으나, 가해자 등 쪽에 정체불명의 기계/금속 물체가 생성된 하드 위반(invented object)이 있으며 지정된 허벅지 상처도 누락되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 5, "verdict_ko": "지정된 미디엄 샷보다 프레이밍이 넓고 인영이 허공에 떠 있지 않으나, 피해자의 허벅지 상처와 신발 한 짝이 벗겨진 상태를 정확히 반영했습니다." }, { "label": "A", "score": 3, "verdict_ko": "허공에서 날아드는 동작은 구현했으나 인영의 등 뒤에 와이어 장치 같은 불필요한 물체가 생성되었고(Hard Violation), 허벅지 상처 누락 및 양발이 맨발로 묘사되어 유지 상태를 어겼습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "지정된 미디엄 샷보다 프레이밍이 넓고 인영이 허공에 떠 있지 않으나, 피해자의 허벅지 상처와 신발 한 짝이 벗겨진 상태를 정확히 반영했습니다." }, { "label": "B", "score": 3, "verdict_ko": "허공에서 날아드는 동작은 구현했으나 인영의 등 뒤에 와이어 장치 같은 불필요한 물체가 생성되었고(Hard Violation), 허벅지 상처 누락 및 양발이 맨발로 묘사되어 유지 상태를 어겼습니다." } ] }, "combined": { "totals": { "A": 10, "B": 6 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S1sh3__bgfirst_bg.png", "bg_asset_id": "22cdf047-8894-4dca-a2a5-41766c746cfc", "bg_record_key": "S1sh3::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "산속 숲 현장", "groupbg_asset_id": "8ad24788-9939-45c0-a31f-5af8b4948512" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "groupbg::도시 뒷골목": { "input_fingerprint": "b1984bd9c50eb0c9", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "도시 뒷골목", "tags": [ "S2sh1", "S2sh4" ] }, "context_sig": "9762c978faf8e3fb" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: Beside a utility pole in a city back alley, outdoors under a dim streetlight at night.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n도시 뒷골목: 도심의 낡고 좁은 골목길로, 낡은 전봇대와 어둡고 로우키 대비가 강한 조명 환경을 갖춘 야간 공간이다. (특징: 희미한 가로등 불빛; 전봇대; 어둡고 거친 낡은 벽면)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#2. 도시 뒷골목 / N\n- 도심 희미한 가로등 불빛. 전봇대에 뒤엉켜있는 두 사람.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: Beside a utility pole in a city back alley, outdoors under a dim streetlight at night.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n도시 뒷골목: 도심의 낡고 좁은 골목길로, 낡은 전봇대와 어둡고 로우키 대비가 강한 조명 환경을 갖춘 야간 공간이다. (특징: 희미한 가로등 불빛; 전봇대; 어둡고 거친 낡은 벽면)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#2. 도시 뒷골목 / N\n- 도심 희미한 가로등 불빛. 전봇대에 뒤엉켜있는 두 사람.\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_도시_뒷골목_0238ec.png", "asset_id": "071e106c-bcbd-40de-84bc-7780bcbb221b", "input_asset_ids": [ "ae97241c-1757-48b3-946f-090d8527e814" ], "origin_tag": "S2sh1", "place_text": "Beside a utility pole in a city back alley, outdoors under a dim streetlight at night.", "origin_inputs": { "place_text": "Beside a utility pole in a city back alley, outdoors under a dim streetlight at night.", "time_of_day_en": "night", "conti_asset_id": "ae97241c-1757-48b3-946f-090d8527e814" } }, "S2sh1::bgfirst_bg": { "input_fingerprint": "e982d1047c998056", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 도시의 뒷골목 전봇대 옆, 희미한 가로등 불빛 아래 한국인 남성과 효장동 피해 여성이 밀착해 있는 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley, outdoors under a dim streetlight at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on the alley axis, the camera begins a slow dolly toward the pair from the alley interior, holding them together in a wide composition beside the utility pole. The man and woman occupy the center-right midground in a compressed embrace, with the pole forming a rigid vertical behind them and the alley remaining visible around their bodies.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 효장동 공격 남성 (활성 상태) in the middle-center of the frame, midground; 효장동 피해 여성 in the middle-right of the frame, midground; 전봇대 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (두 사람 옆에 위치) — Its alley-facing side is visible directly behind the pair; used as Forms the rigid backing against which the pair are intertwined; 도시 뒷골목 (야간); used as Provides spatial depth along the camera's slow approach; 가로등 (켜짐) — The side facing the alley is visible within the setting; used as Provides the explicitly described faint illumination over the pair.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlight provides restrained nocturnal illumination, leaving the surrounding alley in low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 도시의 뒷골목 전봇대 옆, 희미한 가로등 불빛 아래 한국인 남성과 효장동 피해 여성이 밀착해 있는 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley, outdoors under a dim streetlight at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on the alley axis, the camera begins a slow dolly toward the pair from the alley interior, holding them together in a wide composition beside the utility pole. The man and woman occupy the center-right midground in a compressed embrace, with the pole forming a rigid vertical behind them and the alley remaining visible around their bodies.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 효장동 공격 남성 (활성 상태) in the middle-center of the frame, midground; 효장동 피해 여성 in the middle-right of the frame, midground; 전봇대 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (두 사람 옆에 위치) — Its alley-facing side is visible directly behind the pair; used as Forms the rigid backing against which the pair are intertwined; 도시 뒷골목 (야간); used as Provides spatial depth along the camera's slow approach; 가로등 (켜짐) — The side facing the alley is visible within the setting; used as Provides the explicitly described faint illumination over the pair.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlight provides restrained nocturnal illumination, leaving the surrounding alley in low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S2sh1__bgfirst_bg.png", "asset_id": "2a4bc187-b770-4688-a64d-3f696319d818", "input_asset_ids": [ "ae97241c-1757-48b3-946f-090d8527e814", "071e106c-bcbd-40de-84bc-7780bcbb221b" ] }, "S2sh1": { "input_fingerprint": "0147ef0c2f0fbedb", "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): 도시의 뒷골목 전봇대 옆, 희미한 가로등 불빛 아래 한국인 남성과 효장동 피해 여성이 밀착해 있는 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley, outdoors under a dim streetlight at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on the alley axis, the camera begins a slow dolly toward the pair from the alley interior, holding them together in a wide composition beside the utility pole. The man and woman occupy the center-right midground in a compressed embrace, with the pole forming a rigid vertical behind them and the alley remaining visible around their bodies.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 효장동 공격 남성 (활성 상태) in the middle-center of the frame, midground; 효장동 피해 여성 in the middle-right of the frame, midground; 전봇대 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (두 사람 옆에 위치) — Its alley-facing side is visible directly behind the pair; used as Forms the rigid backing against which the pair are intertwined; 도시 뒷골목 (야간); used as Provides spatial depth along the camera's slow approach; 가로등 (켜짐) — The side facing the alley is visible within the setting; used as Provides the explicitly described faint illumination over the pair.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlight provides restrained nocturnal illumination, leaving the surrounding alley in low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 효장동 공격 남성 (활성 상태) (한국인 성인 남성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일); 효장동 피해 여성 (한국인 성인 여성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 도시의 뒷골목 전봇대 옆, 희미한 가로등 불빛 아래 한국인 남성과 효장동 피해 여성이 밀착해 있는 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley, outdoors under a dim streetlight at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on the alley axis, the camera begins a slow dolly toward the pair from the alley interior, holding them together in a wide composition beside the utility pole. The man and woman occupy the center-right midground in a compressed embrace, with the pole forming a rigid vertical behind them and the alley remaining visible around their bodies.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 효장동 공격 남성 (활성 상태) in the middle-center of the frame, midground; 효장동 피해 여성 in the middle-right of the frame, midground; 전봇대 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (두 사람 옆에 위치) — Its alley-facing side is visible directly behind the pair; used as Forms the rigid backing against which the pair are intertwined; 도시 뒷골목 (야간); used as Provides spatial depth along the camera's slow approach; 가로등 (켜짐) — The side facing the alley is visible within the setting; used as Provides the explicitly described faint illumination over the pair.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlight provides restrained nocturnal illumination, leaving the surrounding alley in low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 효장동 공격 남성 (활성 상태) (한국인 성인 남성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일); 효장동 피해 여성 (한국인 성인 여성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 도시의 뒷골목 전봇대 옆, 희미한 가로등 불빛 아래 한국인 남성과 효장동 피해 여성이 밀착해 있는 구도.\n\nLOCATION (lock): Beside a utility pole in a city back alley, outdoors under a dim streetlight at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height on the alley axis, the camera begins a slow dolly toward the pair from the alley interior, holding them together in a wide composition beside the utility pole. The man and woman occupy the center-right midground in a compressed embrace, with the pole forming a rigid vertical behind them and the alley remaining visible around their bodies.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 효장동 공격 남성 (활성 상태) in the middle-center of the frame, midground; 효장동 피해 여성 in the middle-right of the frame, midground; 전봇대 in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 전봇대 (두 사람 옆에 위치) — Its alley-facing side is visible directly behind the pair; used as Forms the rigid backing against which the pair are intertwined; 도시 뒷골목 (야간); used as Provides spatial depth along the camera's slow approach; 가로등 (켜짐) — The side facing the alley is visible within the setting; used as Provides the explicitly described faint illumination over the pair.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Faint streetlight provides restrained nocturnal illumination, leaving the surrounding alley in low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 효장동 공격 남성 (활성 상태) (한국인 성인 남성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일); 효장동 피해 여성 (한국인 성인 여성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S2sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S2sh1.png" }, { "label": "CHARACTER REFERENCE — 효장동 공격 남성 (활성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 효장동 피해 여성: 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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_도시_뒷골목_0238ec.png" }, { "label": "CHARACTER REFERENCE — 효장동 공격 남성 (활성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 효장동 피해 여성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물 배치와 프레이밍, 야간 골목길의 조명과 배경 묘사가 제시된 요구사항을 훌륭하게 충족하며 신체 형태도 안정적입니다." }, { "label": "B", "score": 6, "verdict_ko": "프레이밍과 분위기는 좋으나, 남성의 하체와 여성을 감싼 팔의 해부학적 구조가 다소 어색하게 묘사되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물 배치와 프레이밍, 야간 골목길의 조명과 배경 묘사가 제시된 요구사항을 훌륭하게 충족하며 신체 형태도 안정적입니다." }, { "label": "B", "score": 6, "verdict_ko": "프레이밍과 분위기는 좋으나, 남성의 하체와 여성을 감싼 팔의 해부학적 구조가 다소 어색하게 묘사되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 와이드 샷 구도와 로케이션을 정확히 재현했으며, 인물의 의상과 소품(가방)까지 프레임 내에 충실하게 담아냈습니다." }, { "label": "B", "score": 5, "verdict_ko": "인물 배치와 조명 분위기는 적절하나, 레퍼런스에 없는 노란색 물체가 오른쪽 담장 위에 임의로 생성되어 공간적 일관성을 해쳤습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 와이드 샷 구도와 로케이션을 정확히 재현했으며, 인물의 의상과 소품(가방)까지 프레임 내에 충실하게 담아냈습니다." }, { "label": "A", "score": 5, "verdict_ko": "인물 배치와 조명 분위기는 적절하나, 레퍼런스에 없는 노란색 물체가 오른쪽 담장 위에 임의로 생성되어 공간적 일관성을 해쳤습니다." } ] }, "combined": { "totals": { "A": 12, "B": 13 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 } }, "totals": { "A": 12, "B": 13 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "인물 배치와 프레이밍, 야간 골목길의 조명과 배경 묘사가 제시된 요구사항을 훌륭하게 충족하며 신체 형태도 안정적입니다." }, { "label": "B", "score": 6, "verdict_ko": "프레이밍과 분위기는 좋으나, 남성의 하체와 여성을 감싼 팔의 해부학적 구조가 다소 어색하게 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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_도시_뒷골목_0238ec.png" }, { "label": "CHARACTER REFERENCE — 효장동 공격 남성 (활성 상태): the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 효장동 피해 여성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "여성의 어깨를 감싸고 있는 남성의 왼손이 해부학적으로 심하게 왜곡되어 있습니다.", "fix_en": "Repair the anatomy of the man's left hand resting on the woman's shoulder.", "unfixable": false }, { "issue_ko": "가방을 들고 있는 여성의 오른손 손가락들이 가방 끈과 뭉개져 기괴하게 융합되어 있습니다.", "fix_en": "Redraw the woman's right hand so the fingers clearly and naturally grip the handbag straps.", "unfixable": false }, { "issue_ko": "화면 하단에 보이는 남성의 발에 참조 이미지의 검정 스니커즈가 없고 맨발처럼 묘사되어 있습니다.", "fix_en": "Render the black canvas sneakers with white rubber toe caps on the man's visible feet.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Reduce both figures’ on-screen size together while preserving their embrace and center-right placement, leaving clear space beneath their feet and more visible alley around them.", "issue_ko": "두 인물이 화면 높이의 대부분을 차지하고 발이 하단에 걸려 있어, 지정된 와이드 숏의 중경 인물보다 지나치게 크게 보인다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Repair the anatomy of the man's left hand resting on the woman's shoulder.\n- Redraw the woman's right hand so the fingers clearly and naturally grip the handbag straps.\n- Render the black canvas sneakers with white rubber toe caps on the man's visible feet.\n- Reduce both figures’ on-screen size together while preserving their embrace and center-right placement, leaving clear space beneath their feet and more visible alley around them.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 와이드 샷 구도와 인물 배치를 정확히 따랐으며, 레퍼런스에 있는 남성의 신발까지 온전히 사실적으로 렌더링하여 프롬프트 충실도가 높습니다." }, { "label": "A", "score": 6, "verdict_ko": "전반적인 구도와 배경은 지시사항을 잘 따랐으나, 남성의 발 부분이 불분명하게 렌더링되거나 잘려 있어 사실성과 디테일이 떨어집니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 와이드 샷 구도와 인물 배치를 정확히 따랐으며, 레퍼런스에 있는 남성의 신발까지 온전히 사실적으로 렌더링하여 프롬프트 충실도가 높습니다." }, { "label": "A", "score": 6, "verdict_ko": "전반적인 구도와 배경은 지시사항을 잘 따랐으나, 남성의 발 부분이 불분명하게 렌더링되거나 잘려 있어 사실성과 디테일이 떨어집니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷 구도와 인물의 위치, 배경 및 조명 상태를 전반적으로 충실하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "남성의 발이 렌더링되지 않고 다리가 공중에 떠 있는 심각한 신체 구조 오류가 발생하여 실격입니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 와이드 샷 구도와 인물의 위치, 배경 및 조명 상태를 전반적으로 충실하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "남성의 발이 렌더링되지 않고 다리가 공중에 떠 있는 심각한 신체 구조 오류가 발생하여 실격입니다." } ] }, "combined": { "totals": { "A": 9, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S2sh1__bgfirst_bg.png", "bg_asset_id": "2a4bc187-b770-4688-a64d-3f696319d818", "bg_record_key": "S2sh1::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "도시 뒷골목", "groupbg_asset_id": "071e106c-bcbd-40de-84bc-7780bcbb221b" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S2sh4::variants": { "author_fp": "c566f4f64f068d07", "author": { "variants": [ { "approach_ko": "아스팔트 표면을 스치듯 바라보는 얕은 심도의 밀착 시점으로 신체의 무게와 즉각적인 참상을 강조한다.", "prompt_en": "Photograph from an asphalt-skimming, body-length viewpoint with an intimate, compressed close-up feel. Keep her diagonally fallen upper body dominant in the lower center, with the nearest road grit sharply tactile while focus settles on her slack face, clothing, and bleeding neck wound; let the surrounding asphalt and distant alley details fall rapidly into soft darkness. Use the faint streetlight as a narrow grazing source that catches wet blood, fabric folds, strands of dark hair, and the rough aggregate of the road, leaving the rest in severe low-key shadow." }, { "approach_ko": "조금 더 객관적인 깊은 초점과 절제된 구도로 현장 기록처럼 고요하고 냉혹한 여운을 만든다.", "prompt_en": "Use a restrained, observational close-up from the settled low camera position, with a more neutral lens feel and enough depth to hold both her fully supported upper body and the narrow perimeter of asphalt in clear physical relation. Preserve the lower-center diagonal composition and allow the empty road surface around her to carry visual weight, emphasizing absolute stillness rather than visual distortion. Shape the faint streetlight as a broad, subdued falloff across her body and the ground, retaining detail in the asphalt wear, clothing texture, natural hair, and glossy pooling blood while the alley recedes into controlled black." } ] }, "reused": false }, "S2sh4": { "input_fingerprint": "b0adae7b60c2562a", "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): 피를 흘리며 아스팔트 바닥에 축 늘어진 효장동 피해 여성의 상반신.\n\nLOCATION (lock): The asphalt ground of a city back alley near the utility pole, outdoors beneath faint street lighting. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The descending move settles near the asphalt at the end of its arc, about a body-length from the woman, and tightens around her bloodied upper body. She lies diagonally from the lower-left foreground toward the center, while a narrow margin of asphalt remains around her to preserve the physical aftermath.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 효장동 피해 여성 in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 아스팔트 바닥 (여성이 쓰러져 있음); used as Provides the immediate ground plane around the collapsed upper body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The same faint streetlight falls across the collapsed body with severe low-key contrast and subdued color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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, utility pole, asphalt texture, and nocturnal alley atmosphere from the reference. Preserve the fallen woman's clothing, but exclude the standing man, the embrace, and the earlier upright pose.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Her body is collapsed limply on the asphalt, with her bleeding upper body fully slack against the road.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The Hyojang-dong victim's body is collapsed limply on the asphalt, with her bleeding upper body fully slack against the road and the neck wound left by the attack still bleeding.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 효장동 피해 여성 (한국인 성인 여성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 피를 흘리며 아스팔트 바닥에 축 늘어진 효장동 피해 여성의 상반신.\n\nLOCATION (lock): The asphalt ground of a city back alley near the utility pole, outdoors beneath faint street lighting. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The descending move settles near the asphalt at the end of its arc, about a body-length from the woman, and tightens around her bloodied upper body. She lies diagonally from the lower-left foreground toward the center, while a narrow margin of asphalt remains around her to preserve the physical aftermath.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 효장동 피해 여성 in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 아스팔트 바닥 (여성이 쓰러져 있음); used as Provides the immediate ground plane around the collapsed upper body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The same faint streetlight falls across the collapsed body with severe low-key contrast and subdued color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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, utility pole, asphalt texture, and nocturnal alley atmosphere from the reference. Preserve the fallen woman's clothing, but exclude the standing man, the embrace, and the earlier upright pose.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Her body is collapsed limply on the asphalt, with her bleeding upper body fully slack against the road.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The Hyojang-dong victim's body is collapsed limply on the asphalt, with her bleeding upper body fully slack against the road and the neck wound left by the attack still bleeding.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 효장동 피해 여성 (한국인 성인 여성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 an asphalt-skimming, body-length viewpoint with an intimate, compressed close-up feel. Keep her diagonally fallen upper body dominant in the lower center, with the nearest road grit sharply tactile while focus settles on her slack face, clothing, and bleeding neck wound; let the surrounding asphalt and distant alley details fall rapidly into soft darkness. Use the faint streetlight as a narrow grazing source that catches wet blood, fabric folds, strands of dark hair, and the rough aggregate of the road, leaving the rest in severe low-key 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): 피를 흘리며 아스팔트 바닥에 축 늘어진 효장동 피해 여성의 상반신.\n\nLOCATION (lock): The asphalt ground of a city back alley near the utility pole, outdoors beneath faint street lighting. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The descending move settles near the asphalt at the end of its arc, about a body-length from the woman, and tightens around her bloodied upper body. She lies diagonally from the lower-left foreground toward the center, while a narrow margin of asphalt remains around her to preserve the physical aftermath.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 효장동 피해 여성 in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 아스팔트 바닥 (여성이 쓰러져 있음); used as Provides the immediate ground plane around the collapsed upper body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The same faint streetlight falls across the collapsed body with severe low-key contrast and subdued color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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, utility pole, asphalt texture, and nocturnal alley atmosphere from the reference. Preserve the fallen woman's clothing, but exclude the standing man, the embrace, and the earlier upright pose.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Her body is collapsed limply on the asphalt, with her bleeding upper body fully slack against the road.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The Hyojang-dong victim's body is collapsed limply on the asphalt, with her bleeding upper body fully slack against the road and the neck wound left by the attack still bleeding.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 효장동 피해 여성 (한국인 성인 여성, 평범한 얼굴선, 짙은 검은색 머리카락, 자연스러운 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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, observational close-up from the settled low camera position, with a more neutral lens feel and enough depth to hold both her fully supported upper body and the narrow perimeter of asphalt in clear physical relation. Preserve the lower-center diagonal composition and allow the empty road surface around her to carry visual weight, emphasizing absolute stillness rather than visual distortion. Shape the faint streetlight as a broad, subdued falloff across her body and the ground, retaining detail in the asphalt wear, clothing texture, natural hair, and glossy pooling blood while the alley recedes into controlled black.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "요구된 상반신 클로즈업에 완전히 부합하지는 않지만 B보다 샷 크기 지시에 훨씬 근접했으며, 피를 흘린다는 묘사는 누락되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "명시된 상반신 클로즈업 프레이밍을 완전히 무시하고 전신을 프레임에 담아 샷 크기 우선순위에서 크게 감점되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S2sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 효장동 피해 여성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프롬프트에 명시된 목 부위의 상처와 피 흘리는 모습(바닥의 핏자국 등)이 전혀 묘사되지 않았습니다.", "fix_en": "Add a visible, bleeding wound to her neck and paint wet blood pooling on the asphalt around her upper body.", "unfixable": false }, { "issue_ko": "트렌치코트 안에 입은 상의가 어두운 색상입니다. 레퍼런스 이미지에 맞는 흰색 브이넥 블라우스로 변경해야 합니다.", "fix_en": "Change the dark shirt visible under the trench coat to a white v-neck blouse matching the reference.", "unfixable": false }, { "issue_ko": "상반신 클로즈업 샷이 아니며 인물이 좌측 하단에서 중앙을 향해 대각선으로 눕혀진 구도가 아닙니다. 너무 넓은 앵글로 배경이 과도하게 보입니다.", "fix_en": "Change the camera setup to a tight close-up on the woman's upper body, angling the shot so she lies diagonally from the lower-left foreground, removing the deep alley background.", "unfixable": true } ] }, "composition_critique": { "issues": [ { "fix_en": "Crop tightly around the woman's upper body so it dominates the lower center as a close-up, leaving only a narrow margin of asphalt around her.", "issue_ko": "현재 프레임은 넓은 골목과 전경 아스팔트를 과도하게 포함한 와이드 구도이며, 여성의 상반신이 우측에 치우쳐 계약의 하단 중앙 클로즈업보다 작게 보인다.", "unfixable": false }, { "fix_en": "Minimize the visible directional contradiction by cropping out as much of the right-extending lower body as possible while retaining the upper body.", "issue_ko": "여성의 몸축이 하단 좌측 전경에서 중앙으로 향해야 하지만, 현재는 중앙 부근의 머리에서 우측으로 뻗어 반대 방향으로 배치되어 있다.", "unfixable": true } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add a visible, bleeding wound to her neck and paint wet blood pooling on the asphalt around her upper body.\n- Change the dark shirt visible under the trench coat to a white v-neck blouse matching the reference.\n- Crop tightly around the woman's upper body so it dominates the lower center as a close-up, leaving only a narrow margin of asphalt around her.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "요구된 클로즈업 샷 크기와 카메라 각도를 잘 구현하였으며, 상반신에 집중된 구도와 텍스처 묘사가 프롬프트와 일치합니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 클로즈업 프레이밍을 무시하고 전신이 가깝게 보이는 와이드 샷으로 연출하여 프레임 크기 지침을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "요구된 클로즈업 샷 크기와 카메라 각도를 잘 구현하였으며, 상반신에 집중된 구도와 텍스처 묘사가 프롬프트와 일치합니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 클로즈업 프레이밍을 무시하고 전신이 가깝게 보이는 와이드 샷으로 연출하여 프레임 크기 지침을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "프롬프트에 명시된 클로즈업 샷 크기와 상반신 중심의 대각선 구도를 정확히 구현하였으며, 레퍼런스의 의상 디테일을 잘 유지했습니다." }, { "label": "B", "score": 3, "verdict_ko": "지시된 상반신 클로즈업 대신 전신과 배경이 모두 보이는 와이드 샷을 렌더링하여 프레이밍 지침을 크게 위반했으며, 내부 의상(흰 블라우스)도 일치하지 않습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "프롬프트에 명시된 클로즈업 샷 크기와 상반신 중심의 대각선 구도를 정확히 구현하였으며, 레퍼런스의 의상 디테일을 잘 유지했습니다." }, { "label": "A", "score": 3, "verdict_ko": "지시된 상반신 클로즈업 대신 전신과 배경이 모두 보이는 와이드 샷을 렌더링하여 프레이밍 지침을 크게 위반했으며, 내부 의상(흰 블라우스)도 일치하지 않습니다." } ] }, "combined": { "totals": { "A": 7, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S2sh1" } }, "S3sh1::bgfirst_bg": { "input_fingerprint": "b9679e63c3c40558", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 노란 폴리스 라인이 쳐진 낮의 숲속 현장, 한국인 경찰들이 검은 가방에 담긴 시신을 들것 위로 막 들어 올리는 찰나.\n\nLOCATION (lock): Inside the police cordon at the wooded mountain crime scene, where officers recover a body outdoors in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, the camera holds at chest height outside the police line on a side angle to the stretcher, pausing its lateral track at the instant the bag is lifted. The officers and raised black body bag occupy the central midground, while the police line cuts across the lower foreground and preserves the restricted distance of the news camera.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 폴리스 라인 in the lower-center of the frame, foreground; 들것과 시신 가방 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 노란 폴리스 라인 (설치됨) — Its broad side crosses the camera horizontally; used as Creates a foreground barrier between the news camera and recovery activity; 들것 (시신 가방을 받는 중) — Its long side is angled toward the camera; used as Receives the body bag at the center of the recovery action; 검은 시신 가방 (들것 위로 들어 올려지는 중) — Its side profile is visible as officers raise it; used as Acts as the central recovered evidence without exceeding a natural human scale; 숲속 현장 (낮); used as Places the recovery within the daytime forest scene.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight recorded with restrained broadcast contrast keeps the police activity legible without softening the grim subject.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 노란 폴리스 라인이 쳐진 낮의 숲속 현장, 한국인 경찰들이 검은 가방에 담긴 시신을 들것 위로 막 들어 올리는 찰나.\n\nLOCATION (lock): Inside the police cordon at the wooded mountain crime scene, where officers recover a body outdoors in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, the camera holds at chest height outside the police line on a side angle to the stretcher, pausing its lateral track at the instant the bag is lifted. The officers and raised black body bag occupy the central midground, while the police line cuts across the lower foreground and preserves the restricted distance of the news camera.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 폴리스 라인 in the lower-center of the frame, foreground; 들것과 시신 가방 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 노란 폴리스 라인 (설치됨) — Its broad side crosses the camera horizontally; used as Creates a foreground barrier between the news camera and recovery activity; 들것 (시신 가방을 받는 중) — Its long side is angled toward the camera; used as Receives the body bag at the center of the recovery action; 검은 시신 가방 (들것 위로 들어 올려지는 중) — Its side profile is visible as officers raise it; used as Acts as the central recovered evidence without exceeding a natural human scale; 숲속 현장 (낮); used as Places the recovery within the daytime forest scene.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight recorded with restrained broadcast contrast keeps the police activity legible without softening the grim subject.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S3sh1__bgfirst_bg.png", "asset_id": "ad7d31da-6bab-45c3-95f7-7e2522051b43", "input_asset_ids": [ "fff525bc-df9f-4247-84c4-e4454ca59478", "8ad24788-9939-45c0-a31f-5af8b4948512" ] }, "S3sh1": { "input_fingerprint": "ef65da2be7692967", "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): 노란 폴리스 라인이 쳐진 낮의 숲속 현장, 한국인 경찰들이 검은 가방에 담긴 시신을 들것 위로 막 들어 올리는 찰나.\n\nLOCATION (lock): Inside the police cordon at the wooded mountain crime scene, where officers recover a body outdoors in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, the camera holds at chest height outside the police line on a side angle to the stretcher, pausing its lateral track at the instant the bag is lifted. The officers and raised black body bag occupy the central midground, while the police line cuts across the lower foreground and preserves the restricted distance of the news camera.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 폴리스 라인 in the lower-center of the frame, foreground; 들것과 시신 가방 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 노란 폴리스 라인 (설치됨) — Its broad side crosses the camera horizontally; used as Creates a foreground barrier between the news camera and recovery activity; 들것 (시신 가방을 받는 중) — Its long side is angled toward the camera; used as Receives the body bag at the center of the recovery action; 검은 시신 가방 (들것 위로 들어 올려지는 중) — Its side profile is visible as officers raise it; used as Acts as the central recovered evidence without exceeding a natural human scale; 숲속 현장 (낮); used as Places the recovery within the daytime forest scene.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight recorded with restrained broadcast contrast keeps the police activity legible without softening the grim subject.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): His motionless body is fully enclosed in a black body bag supported on a stretcher as police lift it.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The mountain victim remains dead and his motionless body is fully enclosed in a black body bag supported on a stretcher as police lift it inside the police line.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 노란 폴리스 라인이 쳐진 낮의 숲속 현장, 한국인 경찰들이 검은 가방에 담긴 시신을 들것 위로 막 들어 올리는 찰나.\n\nLOCATION (lock): Inside the police cordon at the wooded mountain crime scene, where officers recover a body outdoors in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, the camera holds at chest height outside the police line on a side angle to the stretcher, pausing its lateral track at the instant the bag is lifted. The officers and raised black body bag occupy the central midground, while the police line cuts across the lower foreground and preserves the restricted distance of the news camera.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 폴리스 라인 in the lower-center of the frame, foreground; 들것과 시신 가방 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 노란 폴리스 라인 (설치됨) — Its broad side crosses the camera horizontally; used as Creates a foreground barrier between the news camera and recovery activity; 들것 (시신 가방을 받는 중) — Its long side is angled toward the camera; used as Receives the body bag at the center of the recovery action; 검은 시신 가방 (들것 위로 들어 올려지는 중) — Its side profile is visible as officers raise it; used as Acts as the central recovered evidence without exceeding a natural human scale; 숲속 현장 (낮); used as Places the recovery within the daytime forest scene.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight recorded with restrained broadcast contrast keeps the police activity legible without softening the grim subject.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): His motionless body is fully enclosed in a black body bag supported on a stretcher as police lift it.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The mountain victim remains dead and his motionless body is fully enclosed in a black body bag supported on a stretcher as police lift it inside the police line.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 노란 폴리스 라인이 쳐진 낮의 숲속 현장, 한국인 경찰들이 검은 가방에 담긴 시신을 들것 위로 막 들어 올리는 찰나.\n\nLOCATION (lock): Inside the police cordon at the wooded mountain crime scene, where officers recover a body outdoors in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Presented as local-news footage, the camera holds at chest height outside the police line on a side angle to the stretcher, pausing its lateral track at the instant the bag is lifted. The officers and raised black body bag occupy the central midground, while the police line cuts across the lower foreground and preserves the restricted distance of the news camera.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 폴리스 라인 in the lower-center of the frame, foreground; 들것과 시신 가방 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 노란 폴리스 라인 (설치됨) — Its broad side crosses the camera horizontally; used as Creates a foreground barrier between the news camera and recovery activity; 들것 (시신 가방을 받는 중) — Its long side is angled toward the camera; used as Receives the body bag at the center of the recovery action; 검은 시신 가방 (들것 위로 들어 올려지는 중) — Its side profile is visible as officers raise it; used as Acts as the central recovered evidence without exceeding a natural human scale; 숲속 현장 (낮); used as Places the recovery within the daytime forest scene.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight recorded with restrained broadcast contrast keeps the police activity legible without softening the grim subject.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): His motionless body is fully enclosed in a black body bag supported on a stretcher as police lift it.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The mountain victim remains dead and his motionless body is fully enclosed in a black body bag supported on a stretcher as police lift it inside the police line.\n\nNo people appear unless the shot text itself says so.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S3sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S3sh1.png" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_산속_숲_현장_3a6a3c.png" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "지정된 뉴스 카메라의 시점을 취하지 않고 화면 안에 카메라맨과 취재진, 과학수사대 등 텍스트에 없는 인물들을 다수 포함하여 프레이밍과 인물 제한 지침을 위반했습니다." }, { "label": "B", "score": 8, "verdict_ko": "지정된 뉴스 카메라의 시점과 프레이밍을 정확히 구현하였으며, 전경의 폴리스 라인 뒤로 경찰관들이 시신 가방을 들것 위로 들어 올리는 순간을 지침대로 잘 표현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "지정된 뉴스 카메라의 시점을 취하지 않고 화면 안에 카메라맨과 취재진, 과학수사대 등 텍스트에 없는 인물들을 다수 포함하여 프레이밍과 인물 제한 지침을 위반했습니다." }, { "label": "B", "score": 8, "verdict_ko": "지정된 뉴스 카메라의 시점과 프레이밍을 정확히 구현하였으며, 전경의 폴리스 라인 뒤로 경찰관들이 시신 가방을 들것 위로 들어 올리는 순간을 지침대로 잘 표현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 뉴스 카메라의 시점과 프레이밍을 정확히 구현했으며, 텍스트에 명시된 경찰관들만 등장시켜 지침을 잘 따랐습니다." }, { "label": "B", "score": 3, "verdict_ko": "지문에 없는 촬영 스태프와 방호복을 입은 인물들을 임의로 추가하여 '명시되지 않은 인물 등장 불가' 지침을 심각하게 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 뉴스 카메라의 시점과 프레이밍을 정확히 구현했으며, 텍스트에 명시된 경찰관들만 등장시켜 지침을 잘 따랐습니다." }, { "label": "A", "score": 3, "verdict_ko": "지문에 없는 촬영 스태프와 방호복을 입은 인물들을 임의로 추가하여 '명시되지 않은 인물 등장 불가' 지침을 심각하게 위반했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 6, "B": 15 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "지정된 뉴스 카메라의 시점을 취하지 않고 화면 안에 카메라맨과 취재진, 과학수사대 등 텍스트에 없는 인물들을 다수 포함하여 프레이밍과 인물 제한 지침을 위반했습니다." }, { "label": "B", "score": 8, "verdict_ko": "지정된 뉴스 카메라의 시점과 프레이밍을 정확히 구현하였으며, 전경의 폴리스 라인 뒤로 경찰관들이 시신 가방을 들것 위로 들어 올리는 순간을 지침대로 잘 표현했습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_산속_숲_현장_3a6a3c.png" } ], "critique": { "issues": [ { "issue_ko": "프롬프트에서 모든 텍스트를 금지했지만, 전경의 폴리스 라인과 경찰관들의 유니폼에 글자가 포함되어 있습니다.", "fix_en": "Remove all text and lettering from the yellow police line and the officers' uniforms, leaving the surfaces plain.", "unfixable": false }, { "issue_ko": "오른쪽 끝에 있는 경찰관이 시신 가방을 쥐고 있는 오른손의 손가락이 기형적으로 길고 구조가 뭉개져 있습니다.", "fix_en": "Reconstruct the rightmost officer's right hand to have normal human anatomy and correctly proportioned fingers gripping the bag.", "unfixable": false }, { "issue_ko": "가방 뒤에 서 있는 왼쪽에서 두 번째 경찰관의 손이 뭉개져 가방과 하나로 융합되어 있습니다.", "fix_en": "Redraw the left-center officer's hands to show anatomically correct fingers that are clearly distinct from the black bag.", "unfixable": false }, { "issue_ko": "왼쪽 배경 숲에 서 있는 경찰관의 얼굴과 몸 형태가 뭉개져 사람의 형체를 제대로 갖추지 못했습니다.", "fix_en": "Restore proper human facial features and bodily anatomy to the background police officer on the far left.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text and lettering from the yellow police line and the officers' uniforms, leaving the surfaces plain.\n- Reconstruct the rightmost officer's right hand to have normal human anatomy and correctly proportioned fingers gripping the bag.\n- Redraw the left-center officer's hands to show anatomically correct fingers that are clearly distinct from the black bag.\n- Restore proper human facial features and bodily anatomy to the background police officer on the far 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.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 앵글과 미드그라운드 배치를 정확히 구현하였으며, 화면 내 텍스트 배제 지시를 충실히 따랐습니다." }, { "label": "A", "score": 5, "verdict_ko": "전체적인 숲속 현장 구도는 적절하나, 텍스트를 금지한 부정 제약 조건을 어기고 폴리스 라인에 글자를 생성했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 앵글과 미드그라운드 배치를 정확히 구현하였으며, 화면 내 텍스트 배제 지시를 충실히 따랐습니다." }, { "label": "A", "score": 5, "verdict_ko": "전체적인 숲속 현장 구도는 적절하나, 텍스트를 금지한 부정 제약 조건을 어기고 폴리스 라인에 글자를 생성했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 앵글과 숏 크기를 정확히 따랐으며, 금지된 텍스트나 워터마크 없이 요구된 모든 요소들을 사실적으로 훌륭하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "전반적인 구도와 배경은 좋으나, 폴리스 라인에 텍스트가 생성되어 '화면 어디에도 텍스트가 없어야 한다'는 지시사항을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 앵글과 숏 크기를 정확히 따랐으며, 금지된 텍스트나 워터마크 없이 요구된 모든 요소들을 사실적으로 훌륭하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "전반적인 구도와 배경은 좋으나, 폴리스 라인에 텍스트가 생성되어 '화면 어디에도 텍스트가 없어야 한다'는 지시사항을 위반했습니다." } ] }, "combined": { "totals": { "A": 9, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S3sh1__bgfirst_bg.png", "bg_asset_id": "ad7d31da-6bab-45c3-95f7-7e2522051b43", "bg_record_key": "S3sh1::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "산속 숲 현장", "groupbg_asset_id": "8ad24788-9939-45c0-a31f-5af8b4948512" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S4sh1::variants": { "author_fp": "3a64139957a075ac", "author": { "variants": [ { "approach_ko": "깊은 초점과 절제된 대칭감으로 동네의 복잡한 지리를 또렷하게 읽히게 하는 지도 같은 부감.", "prompt_en": "Use a clean, expansive aerial perspective with deep focus from edge to edge, making the dense roofscape read almost like a physical map. Organize the surrounding blocks as a restrained web of lines and planes leading inward toward the middle-center, while keeping the destination modest in scale. Crisp directional daylight reveals aged masonry, patched roofing, pipes, and weathered surfaces without romanticizing them; neutral contrast and a slight central visual tension create quiet unease." }, { "approach_ko": "망원 압축감과 겹겹이 포개진 지붕으로 옥탑방이 주변 밀집 건물에 눌린 듯 보이게 하는 부감.", "prompt_en": "Photograph the steep overhead view with a compressed lens feel, stacking roofs and upper walls into dense overlapping layers so the central destination appears hemmed in by its surroundings. Keep the broad establishing scale, but let selective atmospheric softness increase toward the far edges while the middle-center retains firm material detail. Harder sunlit planes alternate with compact pockets of shadow, producing a subdued, watchful mood and emphasizing the neighborhood’s constriction rather than its graphic order." } ] }, "reused": false }, "S4sh1": { "input_fingerprint": "f5ff605be74ec112", "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): 낡은 빌라의 옥탑방과 주변 인천 변두리의 조밀한 건물들이 햇살 아래 넓게 내려다보이는 부감 전경.\n\nLOCATION (lock): An elevated exterior view over an old rooftop room and the tightly packed low-rise buildings of the surrounding suburban neighborhood in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a high aerial position, the camera looks steeply downward across the dense outskirts in a broad establishing composition, beginning a gradual descent without changing its viewing axis. The old villa and rooftop room sit near the center but remain modest within the larger pattern of neighboring buildings, clearly establishing their location rather than isolating them too early.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 오래된 빌라 옥탑방 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 오래된 빌라 옥탑방 (주변 건물 사이에 위치) — Its roof and outward-facing sides are visible from the steep overhead angle; used as Serves as the central destination of the descending aerial move; 인천 변두리의 조밀한 건물들 (조밀하게 배치됨) — Their roofs and upper sides form the surrounding aerial layout; used as Provides the wide urban context surrounding the villa.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight and restrained neutral color keep the aerial geography clear while lending the old rooftop room a subtly uneasy emphasis.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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): 낡은 빌라의 옥탑방과 주변 인천 변두리의 조밀한 건물들이 햇살 아래 넓게 내려다보이는 부감 전경.\n\nLOCATION (lock): An elevated exterior view over an old rooftop room and the tightly packed low-rise buildings of the surrounding suburban neighborhood in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a high aerial position, the camera looks steeply downward across the dense outskirts in a broad establishing composition, beginning a gradual descent without changing its viewing axis. The old villa and rooftop room sit near the center but remain modest within the larger pattern of neighboring buildings, clearly establishing their location rather than isolating them too early.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 오래된 빌라 옥탑방 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 오래된 빌라 옥탑방 (주변 건물 사이에 위치) — Its roof and outward-facing sides are visible from the steep overhead angle; used as Serves as the central destination of the descending aerial move; 인천 변두리의 조밀한 건물들 (조밀하게 배치됨) — Their roofs and upper sides form the surrounding aerial layout; used as Provides the wide urban context surrounding the villa.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight and restrained neutral color keep the aerial geography clear while lending the old rooftop room a subtly uneasy emphasis.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 clean, expansive aerial perspective with deep focus from edge to edge, making the dense roofscape read almost like a physical map. Organize the surrounding blocks as a restrained web of lines and planes leading inward toward the middle-center, while keeping the destination modest in scale. Crisp directional daylight reveals aged masonry, patched roofing, pipes, and weathered surfaces without romanticizing them; neutral contrast and a slight central visual tension create quiet 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): day.\n\nSHOT TEXT (authoritative, Korean): 낡은 빌라의 옥탑방과 주변 인천 변두리의 조밀한 건물들이 햇살 아래 넓게 내려다보이는 부감 전경.\n\nLOCATION (lock): An elevated exterior view over an old rooftop room and the tightly packed low-rise buildings of the surrounding suburban neighborhood in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From a high aerial position, the camera looks steeply downward across the dense outskirts in a broad establishing composition, beginning a gradual descent without changing its viewing axis. The old villa and rooftop room sit near the center but remain modest within the larger pattern of neighboring buildings, clearly establishing their location rather than isolating them too early.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 오래된 빌라 옥탑방 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 오래된 빌라 옥탑방 (주변 건물 사이에 위치) — Its roof and outward-facing sides are visible from the steep overhead angle; used as Serves as the central destination of the descending aerial move; 인천 변두리의 조밀한 건물들 (조밀하게 배치됨) — Their roofs and upper sides form the surrounding aerial layout; used as Provides the wide urban context surrounding the villa.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight and restrained neutral color keep the aerial geography clear while lending the old rooftop room a subtly uneasy emphasis.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 steep overhead view with a compressed lens feel, stacking roofs and upper walls into dense overlapping layers so the central destination appears hemmed in by its surroundings. Keep the broad establishing scale, but let selective atmospheric softness increase toward the far edges while the middle-center retains firm material detail. Harder sunlit planes alternate with compact pockets of shadow, producing a subdued, watchful mood and emphasizing the neighborhood’s constriction rather than its graphic order.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": "지정된 부감 전경의 앵글을 정확히 구현했으며, 기준 사진 속 건물의 핵심 구조(2층과 3층의 돌출된 유리 발코니, 계단)를 충실히 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "부감 앵글은 좋으나 기준 사진의 핵심 구조인 2층 돌출 유리 발코니가 평면 창문으로 잘못 렌더링되어 구조적 일치도가 떨어집니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L11B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/background_chain/seed_bg_rooftop_residence_sel.png" } ], "critique": { "issues": [ { "issue_ko": "중앙 빌라 우측면의 외부 계단이 반대 방향(앞에서 뒤로)으로 올라가며, 옥탑방의 문과 창문 위치 등 전반적인 구조가 지정된 레퍼런스와 다릅니다.", "fix_en": "Rebuild the central villa's right-side external staircase to ascend from back to front, and reposition the rooftop room's door and windows to match the locked structural reference.", "unfixable": true } ] }, "composition_critique": { "issues": [] }, "fix_skipped": true, "fix_skip_reason": "all_issues_unfixable", "ref_mode": "플레이트+seed만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only", "plate_select": { "candidates": { "A": "L11B01", "B": "L11B02", "C": "L11B03", "D": "L11B04" }, "assigned": "L11B01", "choice": "Candidate A", "confident": true, "reason_ko": "샷 텍스트에서 '햇살 아래 넓게 내려다보이는 부감 전경'을 명시하고 있으므로, 유일하게 주간에 촬영된 넓은 하이앵글 뷰를 보여주는 후보 A가 정확히 일치합니다.", "kept": "L11B01" } }, "S5sh2::bgfirst_bg": { "input_fingerprint": "ab77b65b73628d74", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 굳은 표정으로 TV 화면을 뚫어지게 응시하는 강민숙의 상반신.\n\nLOCATION (lock): The compact living and kitchen area inside a rooftop home, lit by a thin shaft of sunlight from its small window and the television screen.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height along the kitchen-to-living-room axis, the tracking frame ends at medium distance in front of Minsook and tightens to her upper body without crossing her television-facing eyeline. She occupies the right-center of frame in three-quarter view, with traces of the active kitchen held softly in the left background and the television remaining just beyond the camera-side edge as her fixed visual anchor.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: TV (뉴스 방송 중) — The active screen faces Minsook but remains just outside the camera-side edge; used as Functions as Minsook's off-screen eyeline anchor and the source of the continuing news report; 싱크대 수도꼭지 (물이 흐르는 중) — Its side profile is visible from the living-room axis; used as Appears in the softened kitchen background as evidence of interrupted preparation; 냄비 (물이 끓는 중이며 김이 피어오름) — Its upper opening is partly visible from the camera position; used as Provides a background cooking action behind Minsook; 거실의 작은 창 (햇살이 가늘게 비침) — Daylight is visible entering through the window into the room; used as Creates a narrow daylight accent within the living room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Thin sunlight from the small living-room window shapes Minsook's dark expression against restrained interior contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 굳은 표정으로 TV 화면을 뚫어지게 응시하는 강민숙의 상반신.\n\nLOCATION (lock): The compact living and kitchen area inside a rooftop home, lit by a thin shaft of sunlight from its small window and the television screen.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height along the kitchen-to-living-room axis, the tracking frame ends at medium distance in front of Minsook and tightens to her upper body without crossing her television-facing eyeline. She occupies the right-center of frame in three-quarter view, with traces of the active kitchen held softly in the left background and the television remaining just beyond the camera-side edge as her fixed visual anchor.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: TV (뉴스 방송 중) — The active screen faces Minsook but remains just outside the camera-side edge; used as Functions as Minsook's off-screen eyeline anchor and the source of the continuing news report; 싱크대 수도꼭지 (물이 흐르는 중) — Its side profile is visible from the living-room axis; used as Appears in the softened kitchen background as evidence of interrupted preparation; 냄비 (물이 끓는 중이며 김이 피어오름) — Its upper opening is partly visible from the camera position; used as Provides a background cooking action behind Minsook; 거실의 작은 창 (햇살이 가늘게 비침) — Daylight is visible entering through the window into the room; used as Creates a narrow daylight accent within the living room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Thin sunlight from the small living-room window shapes Minsook's dark expression against restrained interior contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S5sh2__bgfirst_bg.png", "asset_id": "bed50bba-3264-48d2-a9fe-52d94bf4728d", "input_asset_ids": [ "e0378e8a-fafb-488b-ba47-67b851566f48", "66521554-f627-48a1-a4ef-3434310914f7" ] }, "S5sh2": { "input_fingerprint": "e429d806fde46598", "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): 굳은 표정으로 TV 화면을 뚫어지게 응시하는 강민숙의 상반신.\n\nLOCATION (lock): The compact living and kitchen area inside a rooftop home, lit by a thin shaft of sunlight from its small window and the television screen. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height along the kitchen-to-living-room axis, the tracking frame ends at medium distance in front of Minsook and tightens to her upper body without crossing her television-facing eyeline. She occupies the right-center of frame in three-quarter view, with traces of the active kitchen held softly in the left background and the television remaining just beyond the camera-side edge as her fixed visual anchor.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: TV (뉴스 방송 중) — The active screen faces Minsook but remains just outside the camera-side edge; used as Functions as Minsook's off-screen eyeline anchor and the source of the continuing news report; 싱크대 수도꼭지 (물이 흐르는 중) — Its side profile is visible from the living-room axis; used as Appears in the softened kitchen background as evidence of interrupted preparation; 냄비 (물이 끓는 중이며 김이 피어오름) — Its upper opening is partly visible from the camera position; used as Provides a background cooking action behind Minsook; 거실의 작은 창 (햇살이 가늘게 비침) — Daylight is visible entering through the window into the room; used as Creates a narrow daylight accent within the living room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Thin sunlight from the small living-room window shapes Minsook's dark expression against restrained interior contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook already bears the small circular mark on her wrist; it has been present since long before this moment.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 굳은 표정으로 TV 화면을 뚫어지게 응시하는 강민숙의 상반신.\n\nLOCATION (lock): The compact living and kitchen area inside a rooftop home, lit by a thin shaft of sunlight from its small window and the television screen. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height along the kitchen-to-living-room axis, the tracking frame ends at medium distance in front of Minsook and tightens to her upper body without crossing her television-facing eyeline. She occupies the right-center of frame in three-quarter view, with traces of the active kitchen held softly in the left background and the television remaining just beyond the camera-side edge as her fixed visual anchor.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: TV (뉴스 방송 중) — The active screen faces Minsook but remains just outside the camera-side edge; used as Functions as Minsook's off-screen eyeline anchor and the source of the continuing news report; 싱크대 수도꼭지 (물이 흐르는 중) — Its side profile is visible from the living-room axis; used as Appears in the softened kitchen background as evidence of interrupted preparation; 냄비 (물이 끓는 중이며 김이 피어오름) — Its upper opening is partly visible from the camera position; used as Provides a background cooking action behind Minsook; 거실의 작은 창 (햇살이 가늘게 비침) — Daylight is visible entering through the window into the room; used as Creates a narrow daylight accent within the living room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Thin sunlight from the small living-room window shapes Minsook's dark expression against restrained interior contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook already bears the small circular mark on her wrist; it has been present since long before this moment.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 굳은 표정으로 TV 화면을 뚫어지게 응시하는 강민숙의 상반신.\n\nLOCATION (lock): The compact living and kitchen area inside a rooftop home, lit by a thin shaft of sunlight from its small window and the television screen. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height along the kitchen-to-living-room axis, the tracking frame ends at medium distance in front of Minsook and tightens to her upper body without crossing her television-facing eyeline. She occupies the right-center of frame in three-quarter view, with traces of the active kitchen held softly in the left background and the television remaining just beyond the camera-side edge as her fixed visual anchor.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: TV (뉴스 방송 중) — The active screen faces Minsook but remains just outside the camera-side edge; used as Functions as Minsook's off-screen eyeline anchor and the source of the continuing news report; 싱크대 수도꼭지 (물이 흐르는 중) — Its side profile is visible from the living-room axis; used as Appears in the softened kitchen background as evidence of interrupted preparation; 냄비 (물이 끓는 중이며 김이 피어오름) — Its upper opening is partly visible from the camera position; used as Provides a background cooking action behind Minsook; 거실의 작은 창 (햇살이 가늘게 비침) — Daylight is visible entering through the window into the room; used as Creates a narrow daylight accent within the living room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Thin sunlight from the small living-room window shapes Minsook's dark expression against restrained interior contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook already bears the small circular mark on her wrist; it has been present since long before this moment.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S5sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S5sh2.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "프레이밍이 지정된 미디엄 샷보다 다소 넓지만, 레퍼런스의 공간 구조, 의상, 손목 표식 및 TV의 화면 밖 배치를 충실히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 크기는 적절하나 공간 구조를 완전히 왜곡했고, 지정된 의상을 누락했으며 TV를 오프스크린이 아닌 배경에 잘못 배치했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "프레이밍이 지정된 미디엄 샷보다 다소 넓지만, 레퍼런스의 공간 구조, 의상, 손목 표식 및 TV의 화면 밖 배치를 충실히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 크기는 적절하나 공간 구조를 완전히 왜곡했고, 지정된 의상을 누락했으며 TV를 오프스크린이 아닌 배경에 잘못 배치했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 화면 가장자리의 TV 배치, 로케이션의 공간 구조, 캐릭터 의상 및 손목의 원형 표식을 훌륭하게 구현했으나, 카메라 위치로 인해 주방이 왼쪽이 아닌 오른쪽 배경에 놓였습니다." }, { "label": "A", "score": 3, "verdict_ko": "오프스크린이어야 할 TV가 배경에 등장하여 프레이밍 지시를 어겼으며, 로케이션 사진과 완전히 다른 공간 구조, 일치하지 않는 의상, 잘못된 손목 표식 등 다수의 감점 요인이 있습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 화면 가장자리의 TV 배치, 로케이션의 공간 구조, 캐릭터 의상 및 손목의 원형 표식을 훌륭하게 구현했으나, 카메라 위치로 인해 주방이 왼쪽이 아닌 오른쪽 배경에 놓였습니다." }, { "label": "B", "score": 3, "verdict_ko": "오프스크린이어야 할 TV가 배경에 등장하여 프레이밍 지시를 어겼으며, 로케이션 사진과 완전히 다른 공간 구조, 일치하지 않는 의상, 잘못된 손목 표식 등 다수의 감점 요인이 있습니다." } ] }, "combined": { "totals": { "A": 15, "B": 6 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 15, "B": 6 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "프레이밍이 지정된 미디엄 샷보다 다소 넓지만, 레퍼런스의 공간 구조, 의상, 손목 표식 및 TV의 화면 밖 배치를 충실히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "샷 크기는 적절하나 공간 구조를 완전히 왜곡했고, 지정된 의상을 누락했으며 TV를 오프스크린이 아닌 배경에 잘못 배치했습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B01.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "배경의 싱크대 수도꼭지에서 물이 흐르고 있지 않습니다 (프롬프트 명시 및 배경 참조 이미지와 불일치).", "fix_en": "Add a stream of water flowing down from the sink faucet in the background.", "unfixable": false }, { "issue_ko": "캐릭터의 보라색 셔츠 왼쪽 가슴에 캐릭터 참조 이미지에 있는 로고가 누락되었습니다.", "fix_en": "Add the specific logo from the character reference onto the left chest of the purple polo shirt.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "issue_ko": "선택 사진은 기준 배경과 비교해 카메라 축과 원근이 달라져 창문, 출입문, 주방 설비와 테이블의 화면 위치 및 크기가 변했습니다. 고정된 기준 배경 카메라를 그대로 유지해야 하는 조건을 위반합니다.", "fix_en": "Preserve the current image as much as possible, but match the locked background framing and perspective through cropping and alignment wherever feasible.", "unfixable": true }, { "issue_ko": "민숙의 허벅지와 무릎까지 넓게 보여 상반신에 조여진 미디엄 숏보다 프레이밍이 넓습니다.", "fix_en": "Crop in around Minsook to frame her upper body at medium-shot scale while keeping her in the right-center and preserving her leftward eyeline.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add a stream of water flowing down from the sink faucet in the background.\n- Add the specific logo from the character reference onto the left chest of the purple polo shirt.\n- Crop in around Minsook to frame her upper body at medium-shot scale while keeping her in the right-center and preserving her leftward eyeline.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 상반신(미디엄 샷) 프레이밍을 정확히 따랐으며, 요구된 끓는 냄비와 흐르는 물 등 배경 요소도 충실히 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "상반신만 보여야 하는 미디엄 샷 지시를 어기고 앵글을 넓혀 하반신까지 노출되는 프레이밍 오류가 발생했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 상반신(미디엄 샷) 프레이밍을 정확히 따랐으며, 요구된 끓는 냄비와 흐르는 물 등 배경 요소도 충실히 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "상반신만 보여야 하는 미디엄 샷 지시를 어기고 앵글을 넓혀 하반신까지 노출되는 프레이밍 오류가 발생했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 미디엄 샷(상반신) 구도를 정확히 따랐으며 배경 요소와 손목의 원형 표식을 잘 구현했으나, 의상의 로고가 레퍼런스와 다르게 왜곡되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "지시된 상반신 샷보다 프레임이 너무 넓으며, 화면 밖으로 빠져야 할 TV를 왼쪽 벽면에 임의로 그려 넣어 배경 레퍼런스를 훼손하는 심각한 위반이 있습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 미디엄 샷(상반신) 구도를 정확히 따랐으며 배경 요소와 손목의 원형 표식을 잘 구현했으나, 의상의 로고가 레퍼런스와 다르게 왜곡되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "지시된 상반신 샷보다 프레임이 너무 넓으며, 화면 밖으로 빠져야 할 TV를 왼쪽 벽면에 임의로 그려 넣어 배경 레퍼런스를 훼손하는 심각한 위반이 있습니다." } ] }, "combined": { "totals": { "A": 9, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S5sh2__bgfirst_bg.png", "bg_asset_id": "bed50bba-3264-48d2-a9fe-52d94bf4728d", "bg_record_key": "S5sh2::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B01", "choice": "Candidate A", "confident": true, "reason_ko": "지문에 'TV 화면'을 응시한다고 명시되어 있으므로, 거실에 TV가 배치된 공간이 필요합니다. 현재 배정된 이미지가 이를 잘 보여주므로 유지합니다.", "kept": "L04B01" } }, "S5sh8::variants": { "author_fp": "a7c145096e58b4e5", "author": { "variants": [ { "approach_ko": "얕은 심도로 수리영의 가까운 미소를 선명하게 붙잡고 민숙과 현관을 부드럽게 흐리는 친밀한 시선.", "prompt_en": "Use a gently shallow depth of field that locks onto Suriyoung’s smile at the end of the lateral move, allowing Minsook’s overlapping face and shoulder to fall just slightly softer and the entrance interior to dissolve behind them. Favor soft, diffused interior daylight across the skin, with restrained contrast and natural fabric texture, making the closeness feel tender at first glance while the compressed overlap carries a faint unease." }, { "approach_ko": "두 사람의 겹친 얼굴과 어깨를 함께 또렷하게 유지하고 측면광과 단단한 층위로 미묘한 긴장을 강조하는 시선.", "prompt_en": "Hold both women’s overlapping faces and shoulders in clear focus, using the compact diagonal as a firm graphic line within the close-up while the entrance remains softly legible at the rear. Let directional interior daylight skim laterally across their features, creating delicate tonal separation between foreground and midground; keep the palette cool-neutral and the contrast controlled so Suriyoung’s smile reads clearly against a subtly tense, observational stillness." } ] }, "reused": false }, "S5sh8": { "input_fingerprint": "23409d007985270a", "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): 수리영이 강민숙의 어깨 부근에 얼굴을 바짝 대고 미소 짓는 순간.\n\nLOCATION (lock): The interior entry area of the rooftop home, where shoes are put on beside the front door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height beside Suriyoung, the lateral track settles at its closest point in an intimate two-shot as her face reaches Minsook's shoulder. Suriyoung occupies the left foreground in partial profile while Minsook remains on the right midground, their overlapping faces and shoulders forming a compact diagonal that preserves the affectionate physical contact.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground, moves toward Minsook's shoulder; 강민숙 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 현관 (수리영이 외출 준비 중) — The interior-facing side remains softly visible behind the pair; used as Keeps the affectionate moment anchored near the apartment entrance without competing with the faces.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior daylight supports a gentle facial softness while the restrained palette keeps an undercurrent of unease.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The small circular mark remains on Min-sook’s wrist as Su-ri-young leans close to 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: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일); 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 수리영이 강민숙의 어깨 부근에 얼굴을 바짝 대고 미소 짓는 순간.\n\nLOCATION (lock): The interior entry area of the rooftop home, where shoes are put on beside the front door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height beside Suriyoung, the lateral track settles at its closest point in an intimate two-shot as her face reaches Minsook's shoulder. Suriyoung occupies the left foreground in partial profile while Minsook remains on the right midground, their overlapping faces and shoulders forming a compact diagonal that preserves the affectionate physical contact.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground, moves toward Minsook's shoulder; 강민숙 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 현관 (수리영이 외출 준비 중) — The interior-facing side remains softly visible behind the pair; used as Keeps the affectionate moment anchored near the apartment entrance without competing with the faces.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior daylight supports a gentle facial softness while the restrained palette keeps an undercurrent of unease.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The small circular mark remains on Min-sook’s wrist as Su-ri-young leans close to 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: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일); 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 gently shallow depth of field that locks onto Suriyoung’s smile at the end of the lateral move, allowing Minsook’s overlapping face and shoulder to fall just slightly softer and the entrance interior to dissolve behind them. Favor soft, diffused interior daylight across the skin, with restrained contrast and natural fabric texture, making the closeness feel tender at first glance while the compressed overlap carries a faint 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): day.\n\nSHOT TEXT (authoritative, Korean): 수리영이 강민숙의 어깨 부근에 얼굴을 바짝 대고 미소 짓는 순간.\n\nLOCATION (lock): The interior entry area of the rooftop home, where shoes are put on beside the front door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At shoulder height beside Suriyoung, the lateral track settles at its closest point in an intimate two-shot as her face reaches Minsook's shoulder. Suriyoung occupies the left foreground in partial profile while Minsook remains on the right midground, their overlapping faces and shoulders forming a compact diagonal that preserves the affectionate physical contact.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, foreground, moves toward Minsook's shoulder; 강민숙 in the middle-right of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 현관 (수리영이 외출 준비 중) — The interior-facing side remains softly visible behind the pair; used as Keeps the affectionate moment anchored near the apartment entrance without competing with the faces.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior daylight supports a gentle facial softness while the restrained palette keeps an undercurrent of unease.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The small circular mark remains on Min-sook’s wrist as Su-ri-young leans close to 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: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일); 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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):\nHold both women’s overlapping faces and shoulders in clear focus, using the compact diagonal as a firm graphic line within the close-up while the entrance remains softly legible at the rear. Let directional interior daylight skim laterally across their features, creating delicate tonal separation between foreground and midground; keep the palette cool-neutral and the contrast controlled so Suriyoung’s smile reads clearly against a subtly tense, observational stillness.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 7, "B": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 클로즈업 구도와 두 인물의 배치, 복장을 정확히 따랐으며 이전 샷의 현관 배경을 충실히 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "얼굴이 겹치는 부분의 해부학적 처리가 부자연스러우며, 레퍼런스에 없는 신발장 구조가 배경에 임의로 추가되었습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S5sh2_sel.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "기준 이미지(Previous Shot Still)의 현관문 우측 벽에는 아무것도 없으나, 생성된 이미지에는 어두운 겉옷이 걸려 있습니다.", "fix_en": "Remove the dark garment hanging on the wall to the right of the background door.", "unfixable": false }, { "issue_ko": "강민숙의 보라색 셔츠 로고 내부 디테일이 기준 이미지(Previous Shot Still)와 다릅니다(어두운 원형 배경 누락).", "fix_en": "Fill the inner circle behind the crescent inside the sunburst logo on the purple shirt with a dark color to match the previous shot still.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "issue_ko": "깊이 배치가 반대로 보입니다. 수리영은 왼쪽 중경에서 더 작게 보이고, 강민숙이 오른쪽 전경을 크게 차지하지만 계약은 수리영을 전경, 강민숙을 중경으로 지정합니다.", "fix_en": "Enlarge Suriyoung slightly and reduce Minsook slightly so Suriyoung reads as the left foreground figure and Minsook as the right midground figure, while preserving their current shoulder contact and screen positions.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the dark garment hanging on the wall to the right of the background door.\n- Fill the inner circle behind the crescent inside the sunburst logo on the purple shirt with a dark color to match the previous shot still.\n- Enlarge Suriyoung slightly and reduce Minsook slightly so Suriyoung reads as the left foreground figure and Minsook as the right midground figure, while preserving their current shoulder contact and screen positions.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 앵글과 두 인물의 밀착된 포즈를 훌륭하게 구현했으며, 이전 샷에서 확립된 보라색 셔츠의 흰색 로고 디테일을 정확하게 유지했습니다." }, { "label": "B", "score": 6, "verdict_ko": "구도와 인물의 표정 연기는 프롬프트에 부합하나, 이전 샷에서 흰색이었던 강민숙의 셔츠 로고가 검은색으로 변형되는 오류가 발생했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 앵글과 두 인물의 밀착된 포즈를 훌륭하게 구현했으며, 이전 샷에서 확립된 보라색 셔츠의 흰색 로고 디테일을 정확하게 유지했습니다." }, { "label": "B", "score": 6, "verdict_ko": "구도와 인물의 표정 연기는 프롬프트에 부합하나, 이전 샷에서 흰색이었던 강민숙의 셔츠 로고가 검은색으로 변형되는 오류가 발생했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "프롬프트에 명시된 클로즈업 샷과 두 인물의 다정한 자세를 정확히 연출했으며, 레퍼런스에 없는 임의의 사물을 추가하지 않고 배경 공간을 충실히 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "구도와 인물의 연기는 준수하나, 이전 샷 레퍼런스에는 존재하지 않던 정체불명의 옷(재킷)이 배경 벽에 임의로 생성되어 장소의 일관성을 훼손하는 심각한 위반을 범했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "프롬프트에 명시된 클로즈업 샷과 두 인물의 다정한 자세를 정확히 연출했으며, 레퍼런스에 없는 임의의 사물을 추가하지 않고 배경 공간을 충실히 재현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "구도와 인물의 연기는 준수하나, 이전 샷 레퍼런스에는 존재하지 않던 정체불명의 옷(재킷)이 배경 벽에 임의로 생성되어 장소의 일관성을 훼손하는 심각한 위반을 범했습니다." } ] }, "combined": { "totals": { "A": 12, "B": 13 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S5sh2" } }, "S5sh9::variants": { "author_fp": "8ef6775f26f278c3", "author": { "variants": [ { "approach_ko": "손목을 비스듬히 가로지르는 얕은 심도의 촉각적 인서트로, 햇빛에 걸린 표식과 미소의 흔적을 섬세하게 강조한다.", "prompt_en": "Photograph from wrist height on a close, slightly oblique axis along the forearm, letting the wrist cross the frame on a gentle diagonal with the small round mark held near center. Use very shallow, tactile focus so the mark and the narrow sunlit plane of skin resolve crisply while the surrounding wrist, the slim edge of her lightly smiling lower face, and the entry-area depth fall away softly. Let the incoming daylight graze the skin, producing a restrained highlight on the mark and quiet low-key falloff around it." }, { "approach_ko": "손목 표면을 정면에 가깝게 바라보는 절제된 임상적 인서트로, 중앙의 표식을 또렷하고 담담하게 드러낸다.", "prompt_en": "Frame the insert from wrist height with a near-perpendicular view onto the skin, creating a steady, observational composition with the forearm reading more horizontally and the small round mark near the optical center. Keep enough depth definition for the wrist curvature and the narrow edge of her lightly smiling lower face to remain legible without competing for attention; allow the adjoining living-space window to register only as subdued spatial illumination. Shape the daylight as a narrow clean accent across the mark, with controlled shadow separation that makes its scar-like texture physically convincing rather than graphic." } ] }, "reused": false }, "S5sh9": { "input_fingerprint": "d10fc101879b04c0", "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): 가볍게 웃는 강민숙의 손목 위로 작고 둥근 흉터 같은 표식이 햇빛에 도드라져 보이는 클로즈업.\n\nLOCATION (lock): The entry area just inside the rooftop home, with soft daytime illumination reaching in from the adjoining living space. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The viewing axis lowers from the previous shoulder-level embrace to wrist height and completes the push into an isolated detail of Minsook's forearm. The small round mark sits near frame center and occupies the primary focus, while the surrounding wrist and a narrow edge of her lightly smiling lower face preserve bodily context without allowing the mark to exceed a plausible scale.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 강민숙의 손목 (작고 둥근 흉터 같은 표식이 또렷하게 보임); used as Provides the body surface carrying the revealed mark and the only necessary scale reference; 거실의 작은 창 (햇살이 비침) — Sunlight entering through the window reaches Minsook's wrist; used as Creates the explicit highlight that makes the mark more conspicuous.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: A narrow touch of sunlight catches the round mark while the surrounding skin falls into restrained, low-key separation.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 warm indoor daylight, nearby entry-area materials, and the older woman's unchanged clothing and sleeve from the reference. Preserve the small circular wrist mark, but exclude the younger woman's face and the preceding shoulder-sniffing pose from this wrist 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same long-standing small circular mark remains clearly visible on Min-sook’s wrist.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 가볍게 웃는 강민숙의 손목 위로 작고 둥근 흉터 같은 표식이 햇빛에 도드라져 보이는 클로즈업.\n\nLOCATION (lock): The entry area just inside the rooftop home, with soft daytime illumination reaching in from the adjoining living space. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The viewing axis lowers from the previous shoulder-level embrace to wrist height and completes the push into an isolated detail of Minsook's forearm. The small round mark sits near frame center and occupies the primary focus, while the surrounding wrist and a narrow edge of her lightly smiling lower face preserve bodily context without allowing the mark to exceed a plausible scale.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 강민숙의 손목 (작고 둥근 흉터 같은 표식이 또렷하게 보임); used as Provides the body surface carrying the revealed mark and the only necessary scale reference; 거실의 작은 창 (햇살이 비침) — Sunlight entering through the window reaches Minsook's wrist; used as Creates the explicit highlight that makes the mark more conspicuous.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: A narrow touch of sunlight catches the round mark while the surrounding skin falls into restrained, low-key separation.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 warm indoor daylight, nearby entry-area materials, and the older woman's unchanged clothing and sleeve from the reference. Preserve the small circular wrist mark, but exclude the younger woman's face and the preceding shoulder-sniffing pose from this wrist 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same long-standing small circular mark remains clearly visible on Min-sook’s wrist.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 wrist height on a close, slightly oblique axis along the forearm, letting the wrist cross the frame on a gentle diagonal with the small round mark held near center. Use very shallow, tactile focus so the mark and the narrow sunlit plane of skin resolve crisply while the surrounding wrist, the slim edge of her lightly smiling lower face, and the entry-area depth fall away softly. Let the incoming daylight graze the skin, producing a restrained highlight on the mark and quiet low-key falloff around it.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 가볍게 웃는 강민숙의 손목 위로 작고 둥근 흉터 같은 표식이 햇빛에 도드라져 보이는 클로즈업.\n\nLOCATION (lock): The entry area just inside the rooftop home, with soft daytime illumination reaching in from the adjoining living space. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The viewing axis lowers from the previous shoulder-level embrace to wrist height and completes the push into an isolated detail of Minsook's forearm. The small round mark sits near frame center and occupies the primary focus, while the surrounding wrist and a narrow edge of her lightly smiling lower face preserve bodily context without allowing the mark to exceed a plausible scale.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 강민숙의 손목 (작고 둥근 흉터 같은 표식이 또렷하게 보임); used as Provides the body surface carrying the revealed mark and the only necessary scale reference; 거실의 작은 창 (햇살이 비침) — Sunlight entering through the window reaches Minsook's wrist; used as Creates the explicit highlight that makes the mark more conspicuous.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: A narrow touch of sunlight catches the round mark while the surrounding skin falls into restrained, low-key separation.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 warm indoor daylight, nearby entry-area materials, and the older woman's unchanged clothing and sleeve from the reference. Preserve the small circular wrist mark, but exclude the younger woman's face and the preceding shoulder-sniffing pose from this wrist 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The same long-standing small circular mark remains clearly visible on Min-sook’s wrist.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 insert from wrist height with a near-perpendicular view onto the skin, creating a steady, observational composition with the forearm reading more horizontally and the small round mark near the optical center. Keep enough depth definition for the wrist curvature and the narrow edge of her lightly smiling lower face to remain legible without competing for attention; allow the adjoining living-space window to register only as subdued spatial illumination. Shape the daylight as a narrow clean accent across the mark, with controlled shadow separation that makes its scar-like texture physically convincing rather than graphic.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": "손목 부근에 표식을 정확히 배치하였고, 가볍게 웃는 하관과 햇빛을 받는 디테일이 프롬프트의 클로즈업 앵글과 훌륭하게 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "표식이 손목이 아닌 팔뚝 중앙에 위치해 있으며, 화면 우측 손등의 해부학적 형태가 부자연스럽게 렌더링되었습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S5sh8_sel.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "화면 상단에 보이는 손가락 형태가 뭉개져 기형적으로 렌더링되었습니다.", "fix_en": "Repair the severely deformed fingers at the top of the frame into normal human digits.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Angle the visible forearm into a gentle diagonal across the frame while keeping the round mark near the center and preserving the existing camera and background.", "issue_ko": "팔뚝이 프레임을 거의 수직으로 가로질러, 손목이 완만한 대각선으로 놓여야 한다는 구도 지시와 다릅니다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Repair the severely deformed fingers at the top of the frame into normal human digits.\n- Angle the visible forearm into a gentle diagonal across the frame while keeping the round mark near the center and preserving the existing camera and 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.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 클로즈업 구도와 인물의 미소, 손목의 흉터 표식을 잘 구현했으며 해부학적으로도 무난함." }, { "label": "A", "score": 3, "verdict_ko": "구도와 디테일은 지시사항을 따랐으나, 화면 상단에 보이는 손의 형태가 심하게 일그러진 해부학적 오류가 있음." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 클로즈업 구도와 인물의 미소, 손목의 흉터 표식을 잘 구현했으며 해부학적으로도 무난함." }, { "label": "A", "score": 3, "verdict_ko": "구도와 디테일은 지시사항을 따랐으나, 화면 상단에 보이는 손의 형태가 심하게 일그러진 해부학적 오류가 있음." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 클로즈업 구도, 흉터 표식, 조명 및 인물의 미소를 정확히 연출했으며, 손의 해부학적 구조가 B보다 안정적입니다." }, { "label": "B", "score": 4, "verdict_ko": "구도와 조명 등 전반적인 요소는 잘 반영했으나, 프레임 상단의 손가락 형태가 비정상적으로 뭉개지는 해부학적 오류(하드 위반)가 있습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 클로즈업 구도, 흉터 표식, 조명 및 인물의 미소를 정확히 연출했으며, 손의 해부학적 구조가 B보다 안정적입니다." }, { "label": "A", "score": 4, "verdict_ko": "구도와 조명 등 전반적인 요소는 잘 반영했으나, 프레임 상단의 손가락 형태가 비정상적으로 뭉개지는 해부학적 오류(하드 위반)가 있습니다." } ] }, "combined": { "totals": { "A": 7, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S5sh8" } }, "groupbg::해안 도로": { "input_fingerprint": "d1919882529793f9", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "해안 도로", "tags": [ "S24sh2", "S6sh1", "S9sh1" ] }, "context_sig": "af36e2bbc7435a7a" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: An open stretch of coastal road beside the sea, outdoors in bright daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n서해안 해안도로: 바다와 인접한 해안가 포장도로로, 자전거가 주행할 수 있는 도로폭을 갖추고 있다. (특징: 아스팔트 도로; 바다와 맞닿은 방파제 구조물; 세차게 쏟아지는 비와 젖은 노면(일부 시점))\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#6. 해안가 도로 / D\n- 해안도로를 자전거를 타고 바삐 집으로 달려가는 수리영.\n- 자전거를 타고 빗속을 달리는 수리영.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: An open stretch of coastal road beside the sea, outdoors in bright daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n서해안 해안도로: 바다와 인접한 해안가 포장도로로, 자전거가 주행할 수 있는 도로폭을 갖추고 있다. (특징: 아스팔트 도로; 바다와 맞닿은 방파제 구조물; 세차게 쏟아지는 비와 젖은 노면(일부 시점))\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#6. 해안가 도로 / D\n- 해안도로를 자전거를 타고 바삐 집으로 달려가는 수리영.\n- 자전거를 타고 빗속을 달리는 수리영.\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해안_도로_8394b3.png", "asset_id": "8f0394d6-f50d-4f43-baa3-d2420dc9b871", "input_asset_ids": [ "34691f2e-1752-4b6f-a2ed-6dc4dfc9c935" ], "origin_tag": "S6sh1", "place_text": "An open stretch of coastal road beside the sea, outdoors in bright daylight.", "origin_inputs": { "place_text": "An open stretch of coastal road beside the sea, outdoors in bright daylight.", "time_of_day_en": "day", "conti_asset_id": "34691f2e-1752-4b6f-a2ed-6dc4dfc9c935" } }, "S6sh1::bgfirst_bg": { "input_fingerprint": "a8d0bf83cb9b1fa4", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 밝은 낮, 귀에 이어폰을 꽂은 수리영이 탁 트인 해안가 도로에서 자전거 안장에 앉아 한쪽 페달을 힘껏 내리밟는 중간 자세로, 머리카락과 옷깃이 뒤로 흩날리는 측면 전신.\n\nLOCATION (lock): An open stretch of coastal road beside the sea, outdoors in bright daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height several meters beside Suriyoung, the camera travels parallel with her in a full-body side view as she rides from left to right along the coastal road. Her bicycle and body occupy the middle band of frame, one leg driving the pedal down while the road and open coastal setting remain broad enough to communicate speed and freedom.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward right side of the coastal road.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (밝은 낮) — The road extends along Suriyoung's left-to-right travel direction; used as Defines the horizontal travel axis shared by bicycle and tracking camera; 해안가 (낮); used as Provides the open coastal context behind the moving rider; 자전거 (주행 중) — Its full side profile faces the parallel camera; used as Maintains a clear side silhouette and realistic scale beneath the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Bright daylight gives the moving figure clear separation while maintaining a restrained, natural coastal palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 밝은 낮, 귀에 이어폰을 꽂은 수리영이 탁 트인 해안가 도로에서 자전거 안장에 앉아 한쪽 페달을 힘껏 내리밟는 중간 자세로, 머리카락과 옷깃이 뒤로 흩날리는 측면 전신.\n\nLOCATION (lock): An open stretch of coastal road beside the sea, outdoors in bright daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height several meters beside Suriyoung, the camera travels parallel with her in a full-body side view as she rides from left to right along the coastal road. Her bicycle and body occupy the middle band of frame, one leg driving the pedal down while the road and open coastal setting remain broad enough to communicate speed and freedom.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward right side of the coastal road.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (밝은 낮) — The road extends along Suriyoung's left-to-right travel direction; used as Defines the horizontal travel axis shared by bicycle and tracking camera; 해안가 (낮); used as Provides the open coastal context behind the moving rider; 자전거 (주행 중) — Its full side profile faces the parallel camera; used as Maintains a clear side silhouette and realistic scale beneath the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Bright daylight gives the moving figure clear separation while maintaining a restrained, natural coastal palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S6sh1__bgfirst_bg.png", "asset_id": "5470a5ba-f495-4772-8111-e848c64ea316", "input_asset_ids": [ "34691f2e-1752-4b6f-a2ed-6dc4dfc9c935", "8f0394d6-f50d-4f43-baa3-d2420dc9b871" ] }, "S6sh1": { "input_fingerprint": "16a71bec091996d5", "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): 밝은 낮, 귀에 이어폰을 꽂은 수리영이 탁 트인 해안가 도로에서 자전거 안장에 앉아 한쪽 페달을 힘껏 내리밟는 중간 자세로, 머리카락과 옷깃이 뒤로 흩날리는 측면 전신.\n\nLOCATION (lock): An open stretch of coastal road beside the sea, outdoors in bright daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height several meters beside Suriyoung, the camera travels parallel with her in a full-body side view as she rides from left to right along the coastal road. Her bicycle and body occupy the middle band of frame, one leg driving the pedal down while the road and open coastal setting remain broad enough to communicate speed and freedom.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward right side of the coastal road.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (밝은 낮) — The road extends along Suriyoung's left-to-right travel direction; used as Defines the horizontal travel axis shared by bicycle and tracking camera; 해안가 (낮); used as Provides the open coastal context behind the moving rider; 자전거 (주행 중) — Its full side profile faces the parallel camera; used as Maintains a clear side silhouette and realistic scale beneath the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Bright daylight gives the moving figure clear separation while maintaining a restrained, natural coastal palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young is riding her bicycle and has earphones fitted in her ears.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 밝은 낮, 귀에 이어폰을 꽂은 수리영이 탁 트인 해안가 도로에서 자전거 안장에 앉아 한쪽 페달을 힘껏 내리밟는 중간 자세로, 머리카락과 옷깃이 뒤로 흩날리는 측면 전신.\n\nLOCATION (lock): An open stretch of coastal road beside the sea, outdoors in bright daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height several meters beside Suriyoung, the camera travels parallel with her in a full-body side view as she rides from left to right along the coastal road. Her bicycle and body occupy the middle band of frame, one leg driving the pedal down while the road and open coastal setting remain broad enough to communicate speed and freedom.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward right side of the coastal road.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (밝은 낮) — The road extends along Suriyoung's left-to-right travel direction; used as Defines the horizontal travel axis shared by bicycle and tracking camera; 해안가 (낮); used as Provides the open coastal context behind the moving rider; 자전거 (주행 중) — Its full side profile faces the parallel camera; used as Maintains a clear side silhouette and realistic scale beneath the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Bright daylight gives the moving figure clear separation while maintaining a restrained, natural coastal palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young is riding her bicycle and has earphones fitted in her ears.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 밝은 낮, 귀에 이어폰을 꽂은 수리영이 탁 트인 해안가 도로에서 자전거 안장에 앉아 한쪽 페달을 힘껏 내리밟는 중간 자세로, 머리카락과 옷깃이 뒤로 흩날리는 측면 전신.\n\nLOCATION (lock): An open stretch of coastal road beside the sea, outdoors in bright daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At waist height several meters beside Suriyoung, the camera travels parallel with her in a full-body side view as she rides from left to right along the coastal road. Her bicycle and body occupy the middle band of frame, one leg driving the pedal down while the road and open coastal setting remain broad enough to communicate speed and freedom.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward right side of the coastal road.\n- KEY BACKGROUND ELEMENTS: 해안가 도로 (밝은 낮) — The road extends along Suriyoung's left-to-right travel direction; used as Defines the horizontal travel axis shared by bicycle and tracking camera; 해안가 (낮); used as Provides the open coastal context behind the moving rider; 자전거 (주행 중) — Its full side profile faces the parallel camera; used as Maintains a clear side silhouette and realistic scale beneath the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Bright daylight gives the moving figure clear separation while maintaining a restrained, natural coastal palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young is riding her bicycle and has earphones fitted in her ears.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S6sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S6sh1.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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해안_도로_8394b3.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" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 와이드 샷 프레이밍과 측면 구도를 잘 따랐으며, 위치 레퍼런스의 특징적인 배경(산, 굽은 도로, 펜스)을 정확히 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 자세와 의상은 양호하나, 프레이밍이 다소 좁고 지정된 위치 레퍼런스의 핵심 배경(산, 도로의 형태)이 생략되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 와이드 샷 프레이밍과 측면 구도를 잘 따랐으며, 위치 레퍼런스의 특징적인 배경(산, 굽은 도로, 펜스)을 정확히 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 자세와 의상은 양호하나, 프레이밍이 다소 좁고 지정된 위치 레퍼런스의 핵심 배경(산, 도로의 형태)이 생략되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "인물과 동작의 묘사는 우수하나, 제공된 장소 레퍼런스(곡선 도로, 산맥, 젖은 노면, 특정 난간)를 완전히 무시하고 다른 해안가를 생성하여 위치 제약을 어겼습니다." }, { "label": "B", "score": 7, "verdict_ko": "레퍼런스의 장소(해안가 도로, 산, 노면 상태)를 완벽하게 재현했으며, 이어폰을 꽂고 페달을 밟는 측면 전신 구도와 흩날리는 옷깃까지 지시문 전체를 매우 충실하게 구현했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "인물과 동작의 묘사는 우수하나, 제공된 장소 레퍼런스(곡선 도로, 산맥, 젖은 노면, 특정 난간)를 완전히 무시하고 다른 해안가를 생성하여 위치 제약을 어겼습니다." }, { "label": "A", "score": 7, "verdict_ko": "레퍼런스의 장소(해안가 도로, 산, 노면 상태)를 완벽하게 재현했으며, 이어폰을 꽂고 페달을 밟는 측면 전신 구도와 흩날리는 옷깃까지 지시문 전체를 매우 충실하게 구현했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 8 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 8 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 와이드 샷 프레이밍과 측면 구도를 잘 따랐으며, 위치 레퍼런스의 특징적인 배경(산, 굽은 도로, 펜스)을 정확히 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물의 자세와 의상은 양호하나, 프레이밍이 다소 좁고 지정된 위치 레퍼런스의 핵심 배경(산, 도로의 형태)이 생략되었습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해안_도로_8394b3.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "자전거 체인링 부근에 불필요한 흰색 신발 형태의 잔상이 하나 더 생성되어 있습니다.", "fix_en": "Remove the extra floating white shoe artifact near the bicycle's front chainring.", "unfixable": false }, { "issue_ko": "프롬프트의 '페달을 힘껏 내리밟는 중간 자세'와 달리, 오른쪽 다리가 페달 회전의 가장 높은 곳(12시 방향)에 위치해 있습니다.", "fix_en": "Reposition the right leg, foot, and crank arm to point forward and slightly downward to show a mid-stroke pushing posture.", "unfixable": false }, { "issue_ko": "자전거 체인의 윗부분이 누락되어 뒷바퀴 기어와 앞 체인링이 연결되어 있지 않습니다.", "fix_en": "Draw the missing upper section of the bicycle chain connecting the top of the rear gear to the top of the front chainring.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the extra floating white shoe artifact near the bicycle's front chainring.\n- Reposition the right leg, foot, and crank arm to point forward and slightly downward to show a mid-stroke pushing posture.\n- Draw the missing upper section of the bicycle chain connecting the top of the rear gear to the top of the front chainring.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "바퀴가 지면에서 떠 있고 자전거의 우측 크랭크 축과 체인 연결이 물리적으로 불가능한 형태로 왜곡되어 심각한 규정 위반입니다." }, { "label": "B", "score": 7, "verdict_ko": "정상적인 자전거 구조와 접지 상태를 보여주며, 힘껏 페달을 밟는 자세와 뒤로 흩날리는 옷깃 등 제시된 지시문과 구도를 충실히 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "바퀴가 지면에서 떠 있고 자전거의 우측 크랭크 축과 체인 연결이 물리적으로 불가능한 형태로 왜곡되어 심각한 규정 위반입니다." }, { "label": "B", "score": 7, "verdict_ko": "정상적인 자전거 구조와 접지 상태를 보여주며, 힘껏 페달을 밟는 자세와 뒤로 흩날리는 옷깃 등 제시된 지시문과 구도를 충실히 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "캐릭터의 인상착의와 배경 및 샷 구성은 지시를 잘 따랐으나, 자전거 반대쪽 다리가 렌더링되지 않은 심각한 해부학적 오류가 있습니다." }, { "label": "B", "score": 4, "verdict_ko": "A와 시각적으로 동일한 이미지를 제시하고 있으며, 마찬가지로 반대쪽 다리가 누락되는 해부학적 오류를 포함하고 있습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "캐릭터의 인상착의와 배경 및 샷 구성은 지시를 잘 따랐으나, 자전거 반대쪽 다리가 렌더링되지 않은 심각한 해부학적 오류가 있습니다." }, { "label": "A", "score": 4, "verdict_ko": "A와 시각적으로 동일한 이미지를 제시하고 있으며, 마찬가지로 반대쪽 다리가 누락되는 해부학적 오류를 포함하고 있습니다." } ] }, "combined": { "totals": { "A": 8, "B": 11 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S6sh1__bgfirst_bg.png", "bg_asset_id": "5470a5ba-f495-4772-8111-e848c64ea316", "bg_record_key": "S6sh1::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "해안 도로", "groupbg_asset_id": "8f0394d6-f50d-4f43-baa3-d2420dc9b871" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "groupbg::동물 보호 시설": { "input_fingerprint": "3122304f600a8512", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "동물 보호 시설", "tags": [ "S7sh2", "S7sh5" ] }, "context_sig": "ea9359767a871c58" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: At a caged kennel enclosure within an abandoned-dog shelter, directly in front of the confined dog; the source does not establish whether the enclosure is fully indoors or open-air.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n유기견 보호원: 철창과 진료 공간이 혼재된 동물 보호 시설로, 약간 거칠고 실용적인 구조로 되어 있다. (특징: 철창으로 된 개장; 목욕 및 미용 도구; 낡은 진료대와 주사기; 콘크리트 바닥)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#7. 유기견 보호원 / D\n- 수리영, 갇혀있는 개들 보면서 한숨 후~\n- ‘네’, 하고 달려가 털깎고, 연고 바르고, 목욕시키고등\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: At a caged kennel enclosure within an abandoned-dog shelter, directly in front of the confined dog; the source does not establish whether the enclosure is fully indoors or open-air.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n유기견 보호원: 철창과 진료 공간이 혼재된 동물 보호 시설로, 약간 거칠고 실용적인 구조로 되어 있다. (특징: 철창으로 된 개장; 목욕 및 미용 도구; 낡은 진료대와 주사기; 콘크리트 바닥)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#7. 유기견 보호원 / D\n- 수리영, 갇혀있는 개들 보면서 한숨 후~\n- ‘네’, 하고 달려가 털깎고, 연고 바르고, 목욕시키고등\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_동물_보호_시설_b15368.png", "asset_id": "fb945eb7-f669-4eb0-b261-de24d4a6efde", "input_asset_ids": [ "95694376-778f-4875-92e6-bdf7d13a8227" ], "origin_tag": "S7sh2", "place_text": "At a caged kennel enclosure within an abandoned-dog shelter, directly in front of the confined dog; the source does not establish whether the enclosure is fully indoors or open-air.", "origin_inputs": { "place_text": "At a caged kennel enclosure within an abandoned-dog shelter, directly in front of the confined dog; the source does not establish whether the enclosure is fully indoors or open-air.", "time_of_day_en": "day", "conti_asset_id": "95694376-778f-4875-92e6-bdf7d13a8227" } }, "S7sh2::bgfirst_bg": { "input_fingerprint": "71dbebd1b3dd71b1", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 철창 너머로 로트와일러의 눈을 똑바로 노려보는 수리영의 강렬하고 서늘한 눈빛 클로즈업.\n\nLOCATION (lock): At a caged kennel enclosure within an abandoned-dog shelter, directly in front of the confined dog; the source does not establish whether the enclosure is fully indoors or open-air.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the Rottweiler's side of the enclosure boundary, the camera completes its eye-level track about a meter across from Suriyoung and tightens to the explicitly extreme facial close-up. Her eyes hold the center of frame in a justified confrontational near-frontal angle, while a narrow section of enclosure bars remains at one edge to prove that her stare is directed through the boundary rather than at the audience.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, looks toward Rottweiler across the enclosure boundary; 철창 경계 in the middle-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 철창 (수리영과 로트와일러 사이에 위치) — A small portion of the bar faces the camera edge-on between lens and subject; used as Remains as a narrow edge obstruction that physically separates Suriyoung from the animal.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight is rendered with tight facial contrast and subdued color, concentrating intensity in Suriyoung's unwavering eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 철창 너머로 로트와일러의 눈을 똑바로 노려보는 수리영의 강렬하고 서늘한 눈빛 클로즈업.\n\nLOCATION (lock): At a caged kennel enclosure within an abandoned-dog shelter, directly in front of the confined dog; the source does not establish whether the enclosure is fully indoors or open-air.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the Rottweiler's side of the enclosure boundary, the camera completes its eye-level track about a meter across from Suriyoung and tightens to the explicitly extreme facial close-up. Her eyes hold the center of frame in a justified confrontational near-frontal angle, while a narrow section of enclosure bars remains at one edge to prove that her stare is directed through the boundary rather than at the audience.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, looks toward Rottweiler across the enclosure boundary; 철창 경계 in the middle-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 철창 (수리영과 로트와일러 사이에 위치) — A small portion of the bar faces the camera edge-on between lens and subject; used as Remains as a narrow edge obstruction that physically separates Suriyoung from the animal.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight is rendered with tight facial contrast and subdued color, concentrating intensity in Suriyoung's unwavering eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S7sh2__bgfirst_bg.png", "asset_id": "c3ec44d0-5f0b-4d17-925f-89a3eed904b2", "input_asset_ids": [ "95694376-778f-4875-92e6-bdf7d13a8227", "fb945eb7-f669-4eb0-b261-de24d4a6efde" ] }, "S7sh2": { "input_fingerprint": "1d052ef43f8c0c5a", "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): 철창 너머로 로트와일러의 눈을 똑바로 노려보는 수리영의 강렬하고 서늘한 눈빛 클로즈업.\n\nLOCATION (lock): At a caged kennel enclosure within an abandoned-dog shelter, directly in front of the confined dog; the source does not establish whether the enclosure is fully indoors or open-air. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the Rottweiler's side of the enclosure boundary, the camera completes its eye-level track about a meter across from Suriyoung and tightens to the explicitly extreme facial close-up. Her eyes hold the center of frame in a justified confrontational near-frontal angle, while a narrow section of enclosure bars remains at one edge to prove that her stare is directed through the boundary rather than at the audience.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, looks toward Rottweiler across the enclosure boundary; 철창 경계 in the middle-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 철창 (수리영과 로트와일러 사이에 위치) — A small portion of the bar faces the camera edge-on between lens and subject; used as Remains as a narrow edge obstruction that physically separates Suriyoung from the animal.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight is rendered with tight facial contrast and subdued color, concentrating intensity in Suriyoung's unwavering eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 너머로 로트와일러의 눈을 똑바로 노려보는 수리영의 강렬하고 서늘한 눈빛 클로즈업.\n\nLOCATION (lock): At a caged kennel enclosure within an abandoned-dog shelter, directly in front of the confined dog; the source does not establish whether the enclosure is fully indoors or open-air. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the Rottweiler's side of the enclosure boundary, the camera completes its eye-level track about a meter across from Suriyoung and tightens to the explicitly extreme facial close-up. Her eyes hold the center of frame in a justified confrontational near-frontal angle, while a narrow section of enclosure bars remains at one edge to prove that her stare is directed through the boundary rather than at the audience.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, looks toward Rottweiler across the enclosure boundary; 철창 경계 in the middle-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 철창 (수리영과 로트와일러 사이에 위치) — A small portion of the bar faces the camera edge-on between lens and subject; used as Remains as a narrow edge obstruction that physically separates Suriyoung from the animal.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight is rendered with tight facial contrast and subdued color, concentrating intensity in Suriyoung's unwavering eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 철창 너머로 로트와일러의 눈을 똑바로 노려보는 수리영의 강렬하고 서늘한 눈빛 클로즈업.\n\nLOCATION (lock): At a caged kennel enclosure within an abandoned-dog shelter, directly in front of the confined dog; the source does not establish whether the enclosure is fully indoors or open-air. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the Rottweiler's side of the enclosure boundary, the camera completes its eye-level track about a meter across from Suriyoung and tightens to the explicitly extreme facial close-up. Her eyes hold the center of frame in a justified confrontational near-frontal angle, while a narrow section of enclosure bars remains at one edge to prove that her stare is directed through the boundary rather than at the audience.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, looks toward Rottweiler across the enclosure boundary; 철창 경계 in the middle-left of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 철창 (수리영과 로트와일러 사이에 위치) — A small portion of the bar faces the camera edge-on between lens and subject; used as Remains as a narrow edge obstruction that physically separates Suriyoung from the animal.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight is rendered with tight facial contrast and subdued color, concentrating intensity in Suriyoung's unwavering eyes.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S7sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/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 PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_동물_보호_시설_b15368.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" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 로트와일러 시점의 카메라 위치와 정면 클로즈업 프레이밍, 화면 한쪽의 철창 배치까지 프롬프트의 구도 지시를 정확히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "요구된 정면 카메라 앵글(로트와일러 측면에서 바라본 시점)을 무시하고, 개와 인물을 함께 담는 측면 구도로 임의 변경하여 주요 지침을 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 로트와일러 시점의 카메라 위치와 정면 클로즈업 프레이밍, 화면 한쪽의 철창 배치까지 프롬프트의 구도 지시를 정확히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "요구된 정면 카메라 앵글(로트와일러 측면에서 바라본 시점)을 무시하고, 개와 인물을 함께 담는 측면 구도로 임의 변경하여 주요 지침을 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "로트와일러 시점의 정면 근접 촬영 구도와 왼쪽 철창의 화면 배치, 배경 위치 등 모든 프레이밍 지시를 정확히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "카메라가 철창 바깥쪽에 위치하며 측면 구도로 촬영되어, 로트와일러 측면에서의 정면 근접 촬영이라는 주요 지시사항을 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "로트와일러 시점의 정면 근접 촬영 구도와 왼쪽 철창의 화면 배치, 배경 위치 등 모든 프레이밍 지시를 정확히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "카메라가 철창 바깥쪽에 위치하며 측면 구도로 촬영되어, 로트와일러 측면에서의 정면 근접 촬영이라는 주요 지시사항을 위반했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 7 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 7 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 로트와일러 시점의 카메라 위치와 정면 클로즈업 프레이밍, 화면 한쪽의 철창 배치까지 프롬프트의 구도 지시를 정확히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "요구된 정면 카메라 앵글(로트와일러 측면에서 바라본 시점)을 무시하고, 개와 인물을 함께 담는 측면 구도로 임의 변경하여 주요 지침을 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_동물_보호_시설_b15368.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물이 카메라 렌즈를 정면으로 응시하고 있습니다. 레이아웃 스케치의 화살표 방향과 샷 텍스트에 따라 화면 왼쪽을 강렬하게 노려보도록 시선을 수정해야 합니다.", "fix_en": "Redirect the character's eyes to look sharply toward the middle-left edge of the frame, matching the arrow in the layout sketch.", "unfixable": false }, { "issue_ko": "프롬프트와 레이아웃 스케치에서 철창은 화면 왼쪽 가장자리에 위치해야 하지만, 인물의 오른쪽 눈과 얼굴 중앙을 가로지르는 두껍고 흐릿한 세로 기둥이 잘못 추가되었습니다.", "fix_en": "Remove the thick, out-of-focus vertical pole obstructing the character's right eye, keeping only the cage structure on the left side of the frame.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Limit the visible enclosure boundary to a narrow foreground strip at the left edge, preserving Suriyoung’s position, scale, and the existing viewpoint.", "issue_ko": "철창이 화면 왼쪽 약 3분의 1을 넓게 차지해, 한쪽 가장자리에 좁은 방해물로만 남아야 한다는 구도보다 지나치게 크게 보입니다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redirect the character's eyes to look sharply toward the middle-left edge of the frame, matching the arrow in the layout sketch.\n- Remove the thick, out-of-focus vertical pole obstructing the character's right eye, keeping only the cage structure on the left side of the frame.\n- Limit the visible enclosure boundary to a narrow foreground strip at the left edge, preserving Suriyoung’s position, scale, and the existing viewpoint.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "전경에 철창을 배치하고 카메라를 향해 정면으로 응시하는 시선을 연출하여 프롬프트의 구도와 의도를 정확히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "렌즈와 인물 사이에 있어야 할 전경의 철창이 누락되었으며, 시선이 측면을 향해 정면 대치 구도를 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "전경에 철창을 배치하고 카메라를 향해 정면으로 응시하는 시선을 연출하여 프롬프트의 구도와 의도를 정확히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "렌즈와 인물 사이에 있어야 할 전경의 철창이 누락되었으며, 시선이 측면을 향해 정면 대치 구도를 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "스케치의 시선 방향(왼쪽)과 철창의 가장자리 배치를 정확히 구현하여 렌즈가 아닌 경계 너머를 보는 시선을 잘 표현함." }, { "label": "B", "score": 3, "verdict_ko": "시선이 정면(관객)을 향하고 철창이 얼굴 중앙을 가로막아 레이아웃 스케치 및 프레이밍 지시를 위반함." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "스케치의 시선 방향(왼쪽)과 철창의 가장자리 배치를 정확히 구현하여 렌즈가 아닌 경계 너머를 보는 시선을 잘 표현함." }, { "label": "A", "score": 3, "verdict_ko": "시선이 정면(관객)을 향하고 철창이 얼굴 중앙을 가로막아 레이아웃 스케치 및 프레이밍 지시를 위반함." } ] }, "combined": { "totals": { "A": 10, "B": 10 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S7sh2__bgfirst_bg.png", "bg_asset_id": "c3ec44d0-5f0b-4d17-925f-89a3eed904b2", "bg_record_key": "S7sh2::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "동물 보호 시설", "groupbg_asset_id": "fb945eb7-f669-4eb0-b261-de24d4a6efde" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S7sh5::variants": { "author_fp": "991daf13a3fe75d7", "author": { "variants": [ { "approach_ko": "대각선 금속 날과 양손의 긴장을 또렷하게 분리해 정밀한 작업의 긴장감을 강조한다.", "prompt_en": "Photograph the insert with a restrained, observational precision, keeping the diagonal metal edges, working fingertips, and compressed fur crisply resolved across a compact depth plane. Shape the daytime ambient light from one side so the scissors catch a narrow controlled highlight while the hands and fur retain muted, tactile shadow detail. Use firm geometric balance and subdued contrast to make the suspended instant before the cut feel exacting and tense." }, { "approach_ko": "털의 촉감과 고정한 손의 압력에 초점을 두고 얕은 심도로 친밀하고 불안한 순간을 포착한다.", "prompt_en": "Treat the insert as an intimate tactile close-up with very shallow, natural focus concentrated where the half-open blades approach the fur; let the nearer knuckles and the narrow lap edge fall gently softer without losing their readable form. Keep the camera presence close and quiet, using softly wrapped daytime ambience with minimal metallic flare and delicate tonal separation in the dark neutrals. Emphasize the physical pressure of the steadying hand and the vulnerable softness of the coat, creating hushed apprehension rather than clinical detachment." } ] }, "reused": false }, "S7sh5": { "input_fingerprint": "6467826b65fada5a", "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): 작은 강아지를 무릎에 고정한 채 강아지 털을 향해 가위 날을 반쯤 벌리고 있는 수리영의 양손.\n\nLOCATION (lock): An indoor grooming work area within the dog shelter, set up for trimming and treating the sheltered dogs. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Settle the downward tilt slightly above lap height into an insert that crops 수리영 above the torso, with both hands bracketing the small dog across the lower and middle frame. The half-open scissors cross the dog's fur diagonally without obscuring the restraining hand, while a narrow strip of lap remains as scale context.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 작은 강아지의 털 (Held steady for grooming); used as The fur provides the immediate working surface beneath the open blades; 가위 (Blades partly open) — The blade sides are visible at a diagonal to camera, aimed toward the fur; used as The diagonal focal object between 수리영's hands and the dog's fur; 수리영의 무릎 (Supporting and fixing the dog in place); used as A thin lower-frame scale reference supporting the dog.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daytime ambient light is shaped into restrained low-key contrast with muted neutrals and tactile texture.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 작은 강아지를 무릎에 고정한 채 강아지 털을 향해 가위 날을 반쯤 벌리고 있는 수리영의 양손.\n\nLOCATION (lock): An indoor grooming work area within the dog shelter, set up for trimming and treating the sheltered dogs. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Settle the downward tilt slightly above lap height into an insert that crops 수리영 above the torso, with both hands bracketing the small dog across the lower and middle frame. The half-open scissors cross the dog's fur diagonally without obscuring the restraining hand, while a narrow strip of lap remains as scale context.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 작은 강아지의 털 (Held steady for grooming); used as The fur provides the immediate working surface beneath the open blades; 가위 (Blades partly open) — The blade sides are visible at a diagonal to camera, aimed toward the fur; used as The diagonal focal object between 수리영's hands and the dog's fur; 수리영의 무릎 (Supporting and fixing the dog in place); used as A thin lower-frame scale reference supporting the dog.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daytime ambient light is shaped into restrained low-key contrast with muted neutrals and tactile texture.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 insert with a restrained, observational precision, keeping the diagonal metal edges, working fingertips, and compressed fur crisply resolved across a compact depth plane. Shape the daytime ambient light from one side so the scissors catch a narrow controlled highlight while the hands and fur retain muted, tactile shadow detail. Use firm geometric balance and subdued contrast to make the suspended instant before the cut feel exacting and tense.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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): 작은 강아지를 무릎에 고정한 채 강아지 털을 향해 가위 날을 반쯤 벌리고 있는 수리영의 양손.\n\nLOCATION (lock): An indoor grooming work area within the dog shelter, set up for trimming and treating the sheltered dogs. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Settle the downward tilt slightly above lap height into an insert that crops 수리영 above the torso, with both hands bracketing the small dog across the lower and middle frame. The half-open scissors cross the dog's fur diagonally without obscuring the restraining hand, while a narrow strip of lap remains as scale context.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 작은 강아지의 털 (Held steady for grooming); used as The fur provides the immediate working surface beneath the open blades; 가위 (Blades partly open) — The blade sides are visible at a diagonal to camera, aimed toward the fur; used as The diagonal focal object between 수리영's hands and the dog's fur; 수리영의 무릎 (Supporting and fixing the dog in place); used as A thin lower-frame scale reference supporting the dog.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daytime ambient light is shaped into restrained low-key contrast with muted neutrals and tactile texture.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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):\nTreat the insert as an intimate tactile close-up with very shallow, natural focus concentrated where the half-open blades approach the fur; let the nearer knuckles and the narrow lap edge fall gently softer without losing their readable form. Keep the camera presence close and quiet, using softly wrapped daytime ambience with minimal metallic flare and delicate tonal separation in the dark neutrals. Emphasize the physical pressure of the steadying hand and the vulnerable softness of the coat, creating hushed apprehension rather than clinical detachment.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": "지시된 클로즈업 구도와 양손의 동작을 훌륭하게 구현했으나, 지정된 레퍼런스의 청바지 대신 검은색 바지가 렌더링된 점이 유일한 감점 요소입니다." }, { "label": "A", "score": 3, "verdict_ko": "가위를 쥔 오른손의 손가락과 가위 손잡이가 물리적으로 불가능하게 융합되어 그려진 치명적인 구조적 오류(Hard Violation)로 인해 탈락합니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/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": "화면 하단 무릎 부분의 바지가 검은색으로 렌더링되었습니다. 캐릭터 레퍼런스의 파란색 청바지와 일치해야 합니다.", "fix_en": "Change the color and texture of the pants on the lap from black to faded blue denim.", "unfixable": false }, { "issue_ko": "양손에 핏줄과 주름이 지나치게 두드러져 20대 초반 여성(수리영)의 손으로 보이지 않습니다.", "fix_en": "Smooth the skin texture and reduce the prominent veins on both hands to match a young woman in her early twenties.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Crop tighter around the hands, scissors, dog, and lap to minimize the level-view background and emphasize the intended downward-looking insert.", "issue_ko": "카메라가 무릎보다 약간 높은 위치에서 아래로 기울어진 구도라기보다 강아지와 거의 수평인 축으로 촬영되어, 우측 작업실 배경과 바닥이 지나치게 드러난다.", "unfixable": true } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the color and texture of the pants on the lap from black to faded blue denim.\n- Smooth the skin texture and reduce the prominent veins on both hands to match a young woman in her early twenties.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "레퍼런스 이미지에 제시된 파란색 청바지 복장을 정확히 유지하여 숏의 연속성을 잘 살렸습니다." }, { "label": "A", "score": 6, "verdict_ko": "레퍼런스와 달리 하의가 검은색 바지로 렌더링되어 이전 숏과의 복장 일관성을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "레퍼런스 이미지에 제시된 파란색 청바지 복장을 정확히 유지하여 숏의 연속성을 잘 살렸습니다." }, { "label": "A", "score": 6, "verdict_ko": "레퍼런스와 달리 하의가 검은색 바지로 렌더링되어 이전 숏과의 복장 일관성을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "요구된 인서트 컷의 구도와 동작을 정확히 구현하였으며, 기준 이미지의 청바지 의상을 충실히 반영했습니다." }, { "label": "B", "score": 5, "verdict_ko": "구도와 묘사는 우수하나, 기준 이미지에 명시된 청바지 대신 검은색 바지가 렌더링되어 의상 일관성에서 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "요구된 인서트 컷의 구도와 동작을 정확히 구현하였으며, 기준 이미지의 청바지 의상을 충실히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "구도와 묘사는 우수하나, 기준 이미지에 명시된 청바지 대신 검은색 바지가 렌더링되어 의상 일관성에서 감점되었습니다." } ] }, "combined": { "totals": { "A": 11, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S7sh2" } }, "S8sh2::bgfirst_bg": { "input_fingerprint": "3d9d03b099359617", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 진열대가 늘어선 마트 내부, 수리영이 한국인 중년 여성 직원을 마주 보고 서 있는 구도.\n\nLOCATION (lock): A merchandise aisle inside the large supermarket, between rows of stocked display shelves under store lighting.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level beside the aisle, hold the stage-entry two-shot across the aisle axis before beginning the planned pan, placing 수리영 in the left midground and 한국인 중년 여성 마트 직원 in the right midground. Shelf edges form parallel boundaries behind them while their opposing three-quarter angles keep the exchange readable without either addressing the lens.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged in rows) — The shelf fronts run away from camera on both sides of the aisle, exposing their stocked faces obliquely; used as The rows establish the aisle axis and provide layered depth behind both speakers; 진열대 사이 통로 (Open for passage); used as The clear aisle separates the two figures while preserving their shared conversational space.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient market lighting at dusk is rendered with restrained contrast and a muted-neutral grade without assigning an unsupported color source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 진열대가 늘어선 마트 내부, 수리영이 한국인 중년 여성 직원을 마주 보고 서 있는 구도.\n\nLOCATION (lock): A merchandise aisle inside the large supermarket, between rows of stocked display shelves under store lighting.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level beside the aisle, hold the stage-entry two-shot across the aisle axis before beginning the planned pan, placing 수리영 in the left midground and 한국인 중년 여성 마트 직원 in the right midground. Shelf edges form parallel boundaries behind them while their opposing three-quarter angles keep the exchange readable without either addressing the lens.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged in rows) — The shelf fronts run away from camera on both sides of the aisle, exposing their stocked faces obliquely; used as The rows establish the aisle axis and provide layered depth behind both speakers; 진열대 사이 통로 (Open for passage); used as The clear aisle separates the two figures while preserving their shared conversational space.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient market lighting at dusk is rendered with restrained contrast and a muted-neutral grade without assigning an unsupported color source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S8sh2__bgfirst_bg.png", "asset_id": "578e17e6-f90c-42ab-b829-82c6770e66fe", "input_asset_ids": [ "5e3458bf-dea0-4441-afd5-7cfe44bac8ac", "b0293d55-bc0e-4906-90ec-f4c8dd20141f" ] }, "S8sh2": { "input_fingerprint": "c2209d8d7a1af7a6", "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): 진열대가 늘어선 마트 내부, 수리영이 한국인 중년 여성 직원을 마주 보고 서 있는 구도.\n\nLOCATION (lock): A merchandise aisle inside the large supermarket, between rows of stocked display shelves under store lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level beside the aisle, hold the stage-entry two-shot across the aisle axis before beginning the planned pan, placing 수리영 in the left midground and 한국인 중년 여성 마트 직원 in the right midground. Shelf edges form parallel boundaries behind them while their opposing three-quarter angles keep the exchange readable without either addressing the lens.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged in rows) — The shelf fronts run away from camera on both sides of the aisle, exposing their stocked faces obliquely; used as The rows establish the aisle axis and provide layered depth behind both speakers; 진열대 사이 통로 (Open for passage); used as The clear aisle separates the two figures while preserving their shared conversational space.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient market lighting at dusk is rendered with restrained contrast and a muted-neutral grade without assigning an unsupported color source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s bicycle remains parked outside the mart while she speaks to the employee inside.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 진열대가 늘어선 마트 내부, 수리영이 한국인 중년 여성 직원을 마주 보고 서 있는 구도.\n\nLOCATION (lock): A merchandise aisle inside the large supermarket, between rows of stocked display shelves under store lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level beside the aisle, hold the stage-entry two-shot across the aisle axis before beginning the planned pan, placing 수리영 in the left midground and 한국인 중년 여성 마트 직원 in the right midground. Shelf edges form parallel boundaries behind them while their opposing three-quarter angles keep the exchange readable without either addressing the lens.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged in rows) — The shelf fronts run away from camera on both sides of the aisle, exposing their stocked faces obliquely; used as The rows establish the aisle axis and provide layered depth behind both speakers; 진열대 사이 통로 (Open for passage); used as The clear aisle separates the two figures while preserving their shared conversational space.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient market lighting at dusk is rendered with restrained contrast and a muted-neutral grade without assigning an unsupported color source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s bicycle remains parked outside the mart while she speaks to the employee inside.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 진열대가 늘어선 마트 내부, 수리영이 한국인 중년 여성 직원을 마주 보고 서 있는 구도.\n\nLOCATION (lock): A merchandise aisle inside the large supermarket, between rows of stocked display shelves under store lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level beside the aisle, hold the stage-entry two-shot across the aisle axis before beginning the planned pan, placing 수리영 in the left midground and 한국인 중년 여성 마트 직원 in the right midground. Shelf edges form parallel boundaries behind them while their opposing three-quarter angles keep the exchange readable without either addressing the lens.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 진열대 (Arranged in rows) — The shelf fronts run away from camera on both sides of the aisle, exposing their stocked faces obliquely; used as The rows establish the aisle axis and provide layered depth behind both speakers; 진열대 사이 통로 (Open for passage); used as The clear aisle separates the two figures while preserving their shared conversational space.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient market lighting at dusk is rendered with restrained contrast and a muted-neutral grade without assigning an unsupported color source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s bicycle remains parked outside the mart while she speaks to the employee inside.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S8sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S8sh2.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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L08B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷 크기를 잘 구현하여 인물들의 전신과 배경을 정확하게 담아냈으며, 마트 배경과 캐릭터 디자인이 레퍼런스와 일치합니다." }, { "label": "B", "score": 5, "verdict_ko": "요청된 와이드 샷이 아닌 무릎 아래가 잘린 프레이밍을 사용하여 샷 크기 지침을 어겼습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷 크기를 잘 구현하여 인물들의 전신과 배경을 정확하게 담아냈으며, 마트 배경과 캐릭터 디자인이 레퍼런스와 일치합니다." }, { "label": "B", "score": 5, "verdict_ko": "요청된 와이드 샷이 아닌 무릎 아래가 잘린 프레이밍을 사용하여 샷 크기 지침을 어겼습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 와이드 샷에 맞게 인물의 전신과 배경을 적절한 스케일로 담아내어 구도와 거리감을 정확히 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "요구된 와이드 샷에 비해 인물이 프레임에 크게 잡혀 샷 사이즈 지침을 완벽히 충족하지 못했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷에 맞게 인물의 전신과 배경을 적절한 스케일로 담아내어 구도와 거리감을 정확히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "요구된 와이드 샷에 비해 인물이 프레임에 크게 잡혀 샷 사이즈 지침을 완벽히 충족하지 못했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 10 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 10 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷 크기를 잘 구현하여 인물들의 전신과 배경을 정확하게 담아냈으며, 마트 배경과 캐릭터 디자인이 레퍼런스와 일치합니다." }, { "label": "B", "score": 5, "verdict_ko": "요청된 와이드 샷이 아닌 무릎 아래가 잘린 프레이밍을 사용하여 샷 크기 지침을 어겼습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L08B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [] }, "composition_critique": { "issues": [ { "issue_ko": "두 인물이 레이아웃 스케치보다 화면에서 작아 발밑 바닥 여백까지 보이며, 스케치에서 요구한 하단 프레임까지 차는 인물 크기에 미치지 못한다.", "fix_en": "Enlarge both figures by about 10% in place, anchoring them downward so their feet reach the bottom frame edge while preserving their depth and spacing.", "unfixable": false }, { "issue_ko": "수리영이 스케치의 좌측 위치보다 화면 중앙 쪽에 치우쳐 있다.", "fix_en": "Shift Su-ri-young left by about 5% of the frame width while preserving her scale, depth, and gaze toward the employee.", "unfixable": false }, { "issue_ko": "두 인물이 서로 마주 보는 거의 완전한 측면 자세여서, 계약이 요구한 서로 대응하는 삼사분면 각도가 충분히 드러나지 않는다.", "fix_en": "Turn both figures' torsos and heads slightly toward the camera into opposing three-quarter angles while keeping their eyes directed at each other.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Enlarge both figures by about 10% in place, anchoring them downward so their feet reach the bottom frame edge while preserving their depth and spacing.\n- Shift Su-ri-young left by about 5% of the frame width while preserving her scale, depth, and gaze toward the employee.\n- Turn both figures' torsos and heads slightly toward the camera into opposing three-quarter angles while keeping their eyes directed at each other.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 배경과 레이아웃을 정확하게 유지했으며, 지시된 구도 내에서 인물의 비율, 자세, 표정이 자연스럽게 구현되었습니다." }, { "label": "B", "score": 6, "verdict_ko": "구도와 배경 유지는 훌륭하나, 인물의 손과 발, 그리고 얼굴 형태의 해부학적 디테일에서 미세한 어색함이 관찰됩니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 배경과 레이아웃을 정확하게 유지했으며, 지시된 구도 내에서 인물의 비율, 자세, 표정이 자연스럽게 구현되었습니다." }, { "label": "B", "score": 6, "verdict_ko": "구도와 배경 유지는 훌륭하나, 인물의 손과 발, 그리고 얼굴 형태의 해부학적 디테일에서 미세한 어색함이 관찰됩니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 배경과 스케치의 구도를 완벽하게 유지했으며, 수리영의 외모와 헤어스타일이 캐릭터 레퍼런스와 가장 일치합니다." }, { "label": "A", "score": 6, "verdict_ko": "배경과 구도 지침을 잘 따랐으나, 수리영의 헤어스타일이 레퍼런스와 다소 차이가 있어 아쉽습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 배경과 스케치의 구도를 완벽하게 유지했으며, 수리영의 외모와 헤어스타일이 캐릭터 레퍼런스와 가장 일치합니다." }, { "label": "B", "score": 6, "verdict_ko": "배경과 구도 지침을 잘 따랐으나, 수리영의 헤어스타일이 레퍼런스와 다소 차이가 있어 아쉽습니다." } ] }, "combined": { "totals": { "A": 14, "B": 12 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S8sh2__bgfirst_bg.png", "bg_asset_id": "578e17e6-f90c-42ab-b829-82c6770e66fe", "bg_record_key": "S8sh2::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L08B01", "B": "L08B02" }, "assigned": "L08B01", "choice": "Candidate A", "confident": true, "reason_ko": "\"진열대가 늘어선 마트 내부\"라는 설명과 일치하는 실내 컷이 필요하므로 현재 설정된 후보가 적합합니다.", "kept": "L08B01" } }, "S8sh4::bgfirst_bg": { "input_fingerprint": "fb3bc001cba8f181", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 마트 밖으로 나온 수리영이 귀에 휴대폰을 댄 채 어두워진 길거리를 주시하는 모습.\n\nLOCATION (lock): On the street immediately outside the large supermarket entrance, amid the darkening commercial district at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level in front of and slightly to one side of 수리영, tighten to a medium frame while keeping the darkened street extending through the side of the composition she watches. She occupies the left half with the phone pressed to her ear, leaving the right background open as an uneasy continuation of her sightline.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward darkened street; darkened street in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 어두워진 길거리 (Darkened at dusk); used as Preserved as negative space in 수리영's viewing direction; 마트 외부 (수리영 has exited the market) — The entrance side faces the camera obliquely behind 수리영; used as Provides recognizable exterior context beyond 수리영 without becoming the focal plane.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading dusk ambience and the darkened street are held in low-key contrast with restrained highlights and muted color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 마트 밖으로 나온 수리영이 귀에 휴대폰을 댄 채 어두워진 길거리를 주시하는 모습.\n\nLOCATION (lock): On the street immediately outside the large supermarket entrance, amid the darkening commercial district at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level in front of and slightly to one side of 수리영, tighten to a medium frame while keeping the darkened street extending through the side of the composition she watches. She occupies the left half with the phone pressed to her ear, leaving the right background open as an uneasy continuation of her sightline.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward darkened street; darkened street in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 어두워진 길거리 (Darkened at dusk); used as Preserved as negative space in 수리영's viewing direction; 마트 외부 (수리영 has exited the market) — The entrance side faces the camera obliquely behind 수리영; used as Provides recognizable exterior context beyond 수리영 without becoming the focal plane.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading dusk ambience and the darkened street are held in low-key contrast with restrained highlights and muted color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S8sh4__bgfirst_bg.png", "asset_id": "8eca7f07-3cb6-4aa9-ae7c-cc4203b2c020", "input_asset_ids": [ "06429d9d-4fce-4760-9754-6d9da658ce19", "3853a5af-11f6-4994-90b4-b4ee6dd7c190" ] }, "S8sh4": { "input_fingerprint": "bf7eb61e2882c08f", "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): 마트 밖으로 나온 수리영이 귀에 휴대폰을 댄 채 어두워진 길거리를 주시하는 모습.\n\nLOCATION (lock): On the street immediately outside the large supermarket entrance, amid the darkening commercial district at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level in front of and slightly to one side of 수리영, tighten to a medium frame while keeping the darkened street extending through the side of the composition she watches. She occupies the left half with the phone pressed to her ear, leaving the right background open as an uneasy continuation of her sightline.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward darkened street; darkened street in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 어두워진 길거리 (Darkened at dusk); used as Preserved as negative space in 수리영's viewing direction; 마트 외부 (수리영 has exited the market) — The entrance side faces the camera obliquely behind 수리영; used as Provides recognizable exterior context beyond 수리영 without becoming the focal plane.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading dusk ambience and the darkened street are held in low-key contrast with restrained highlights and muted color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young has her phone at her ear, and her bicycle remains available outside for her trip home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 마트 밖으로 나온 수리영이 귀에 휴대폰을 댄 채 어두워진 길거리를 주시하는 모습.\n\nLOCATION (lock): On the street immediately outside the large supermarket entrance, amid the darkening commercial district at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level in front of and slightly to one side of 수리영, tighten to a medium frame while keeping the darkened street extending through the side of the composition she watches. She occupies the left half with the phone pressed to her ear, leaving the right background open as an uneasy continuation of her sightline.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward darkened street; darkened street in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 어두워진 길거리 (Darkened at dusk); used as Preserved as negative space in 수리영's viewing direction; 마트 외부 (수리영 has exited the market) — The entrance side faces the camera obliquely behind 수리영; used as Provides recognizable exterior context beyond 수리영 without becoming the focal plane.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading dusk ambience and the darkened street are held in low-key contrast with restrained highlights and muted color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young has her phone at her ear, and her bicycle remains available outside for her trip home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 마트 밖으로 나온 수리영이 귀에 휴대폰을 댄 채 어두워진 길거리를 주시하는 모습.\n\nLOCATION (lock): On the street immediately outside the large supermarket entrance, amid the darkening commercial district at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level in front of and slightly to one side of 수리영, tighten to a medium frame while keeping the darkened street extending through the side of the composition she watches. She occupies the left half with the phone pressed to her ear, leaving the right background open as an uneasy continuation of her sightline.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, looks toward darkened street; darkened street in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 어두워진 길거리 (Darkened at dusk); used as Preserved as negative space in 수리영's viewing direction; 마트 외부 (수리영 has exited the market) — The entrance side faces the camera obliquely behind 수리영; used as Provides recognizable exterior context beyond 수리영 without becoming the focal plane.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fading dusk ambience and the darkened street are held in low-key contrast with restrained highlights and muted color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young has her phone at her ear, and her bicycle remains available outside for her trip home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S8sh4__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S8sh4.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L08B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 미디엄 샷과 좌측 배치, 그리고 우측의 어두운 거리를 향하는 시선 방향을 완벽히 구현했으며 레퍼런스의 배경 장소를 정확하게 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "피사체가 중앙에 위치하고 시선이 지시된 우측 배경(어두운 거리)을 향하지 않아 프레이밍 규칙을 위반했으며, 배경 건물 구조가 레퍼런스와 다릅니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 미디엄 샷과 좌측 배치, 그리고 우측의 어두운 거리를 향하는 시선 방향을 완벽히 구현했으며 레퍼런스의 배경 장소를 정확하게 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "피사체가 중앙에 위치하고 시선이 지시된 우측 배경(어두운 거리)을 향하지 않아 프레이밍 규칙을 위반했으며, 배경 건물 구조가 레퍼런스와 다릅니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "지정된 미디엄 샷 크기와 인물의 화면 좌측 배치를 완벽하게 구현하여 프레이밍 지침을 정확히 따랐습니다." }, { "label": "A", "score": 7, "verdict_ko": "인물과 배경의 구현은 훌륭하나, 인물이 화면 중앙에 배치되고 샷이 다소 넓어 '좌측 배치' 및 '미디엄 샷' 지침에서 감점되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 미디엄 샷 크기와 인물의 화면 좌측 배치를 완벽하게 구현하여 프레이밍 지침을 정확히 따랐습니다." }, { "label": "B", "score": 7, "verdict_ko": "인물과 배경의 구현은 훌륭하나, 인물이 화면 중앙에 배치되고 샷이 다소 넓어 '좌측 배치' 및 '미디엄 샷' 지침에서 감점되었습니다." } ] }, "combined": { "totals": { "A": 18, "B": 11 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 18, "B": 11 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 미디엄 샷과 좌측 배치, 그리고 우측의 어두운 거리를 향하는 시선 방향을 완벽히 구현했으며 레퍼런스의 배경 장소를 정확하게 재현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "피사체가 중앙에 위치하고 시선이 지시된 우측 배경(어두운 거리)을 향하지 않아 프레이밍 규칙을 위반했으며, 배경 건물 구조가 레퍼런스와 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L08B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "배경 원본에 있던 자전거가 사라졌으며, 자전거가 밖에 남아있어야 한다는 지시사항을 위반했습니다.", "fix_en": "Restore the silver bicycle to the parking rack behind the character, matching the background reference image exactly.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the silver bicycle to the parking rack behind the character, matching the background reference image exactly.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "스케치의 배경 선을 무시하라는 지시를 준수하여 원본 배경을 최대한 유지했으며, 인물의 위치, 포즈, 외양을 제시된 텍스트와 레퍼런스에 맞게 충실히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "스케치의 배경 선(자전거)을 무시하지 않고 그대로 렌더링하여, 원본 배경을 전혀 수정 없이 유지하라는 핵심 지시사항을 심각하게 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "스케치의 배경 선을 무시하라는 지시를 준수하여 원본 배경을 최대한 유지했으며, 인물의 위치, 포즈, 외양을 제시된 텍스트와 레퍼런스에 맞게 충실히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "스케치의 배경 선(자전거)을 무시하지 않고 그대로 렌더링하여, 원본 배경을 전혀 수정 없이 유지하라는 핵심 지시사항을 심각하게 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "원본 배경의 자전거를 온전히 유지했으며, 요구된 프레이밍과 인물의 포즈를 충실하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "유지되어야 할 배경의 자전거가 지워진 심각한 오류가 있어 지시 사항을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "원본 배경의 자전거를 온전히 유지했으며, 요구된 프레이밍과 인물의 포즈를 충실하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "유지되어야 할 배경의 자전거가 지워진 심각한 오류가 있어 지시 사항을 위반했습니다." } ] }, "combined": { "totals": { "A": 10, "B": 12 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S8sh4__bgfirst_bg.png", "bg_asset_id": "8eca7f07-3cb6-4aa9-ae7c-cc4203b2c020", "bg_record_key": "S8sh4::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L08B01", "B": "L08B02" }, "assigned": "L08B02", "choice": "Candidate B", "confident": true, "reason_ko": "지문에서 '마트 밖으로 나온', '어두워진 길거리'를 명시하므로 마트 외부의 어두운 길거리가 보이는 Candidate B가 적합합니다.", "kept": "L08B02" } }, "S9sh1::bgfirst_bg": { "input_fingerprint": "85238bd654cb512c", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 밤, 푸르스름한 가로등 불빛 아래 해안도로에서 수리영이 상체를 웅크린 채 자전거 페달에 체중을 싣는 찰나.\n\nLOCATION (lock): A nighttime stretch of coastal road beside the sea, outdoors under sparse bluish streetlights.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at bicycle-frame height from 수리영's front-left, tightening enough to bind her hunched upper body and the descending pedal stroke into one moving medium composition. Her torso occupies the upper center while the weight-bearing leg and pedal remain visible in the lower right, with the coastal road sliding laterally behind her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead; road ahead in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안도로 (Dark at night); used as Carries 수리영 through the frame and supplies lateral motion behind the bicycle; 가로등 (On) — The lit streetlamp is positioned above the roadway, outside or near the upper frame boundary; used as Provides the stated blue illumination over the moving rider; 자전거 (In motion) — Its front-left side is presented to camera while traveling parallel to the lens; used as Keeps the pedal mechanism and frame visible as scale and action references.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Blue streetlamp light cuts across the night scene in moody low-key contrast, leaving the surrounding palette subdued.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 밤, 푸르스름한 가로등 불빛 아래 해안도로에서 수리영이 상체를 웅크린 채 자전거 페달에 체중을 싣는 찰나.\n\nLOCATION (lock): A nighttime stretch of coastal road beside the sea, outdoors under sparse bluish streetlights.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at bicycle-frame height from 수리영's front-left, tightening enough to bind her hunched upper body and the descending pedal stroke into one moving medium composition. Her torso occupies the upper center while the weight-bearing leg and pedal remain visible in the lower right, with the coastal road sliding laterally behind her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead; road ahead in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안도로 (Dark at night); used as Carries 수리영 through the frame and supplies lateral motion behind the bicycle; 가로등 (On) — The lit streetlamp is positioned above the roadway, outside or near the upper frame boundary; used as Provides the stated blue illumination over the moving rider; 자전거 (In motion) — Its front-left side is presented to camera while traveling parallel to the lens; used as Keeps the pedal mechanism and frame visible as scale and action references.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Blue streetlamp light cuts across the night scene in moody low-key contrast, leaving the surrounding palette subdued.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S9sh1__bgfirst_bg.png", "asset_id": "31c7a969-a098-46d3-a284-6704e6eb6834", "input_asset_ids": [ "3c8094b8-4dc5-4ee3-b63b-fe4bebb0f3db", "8f0394d6-f50d-4f43-baa3-d2420dc9b871" ] }, "S9sh1": { "input_fingerprint": "7e27fd697fabb010", "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 밤, 푸르스름한 가로등 불빛 아래 해안도로에서 수리영이 상체를 웅크린 채 자전거 페달에 체중을 싣는 찰나.\n\nLOCATION (lock): A nighttime stretch of coastal road beside the sea, outdoors under sparse bluish streetlights. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at bicycle-frame height from 수리영's front-left, tightening enough to bind her hunched upper body and the descending pedal stroke into one moving medium composition. Her torso occupies the upper center while the weight-bearing leg and pedal remain visible in the lower right, with the coastal road sliding laterally behind her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead; road ahead in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안도로 (Dark at night); used as Carries 수리영 through the frame and supplies lateral motion behind the bicycle; 가로등 (On) — The lit streetlamp is positioned above the roadway, outside or near the upper frame boundary; used as Provides the stated blue illumination over the moving rider; 자전거 (In motion) — Its front-left side is presented to camera while traveling parallel to the lens; used as Keeps the pedal mechanism and frame visible as scale and action references.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Blue streetlamp light cuts across the night scene in moody low-key contrast, leaving the surrounding palette subdued.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young is still using the same bicycle as she hurries home along the coastal road.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤, 푸르스름한 가로등 불빛 아래 해안도로에서 수리영이 상체를 웅크린 채 자전거 페달에 체중을 싣는 찰나.\n\nLOCATION (lock): A nighttime stretch of coastal road beside the sea, outdoors under sparse bluish streetlights. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at bicycle-frame height from 수리영's front-left, tightening enough to bind her hunched upper body and the descending pedal stroke into one moving medium composition. Her torso occupies the upper center while the weight-bearing leg and pedal remain visible in the lower right, with the coastal road sliding laterally behind her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead; road ahead in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안도로 (Dark at night); used as Carries 수리영 through the frame and supplies lateral motion behind the bicycle; 가로등 (On) — The lit streetlamp is positioned above the roadway, outside or near the upper frame boundary; used as Provides the stated blue illumination over the moving rider; 자전거 (In motion) — Its front-left side is presented to camera while traveling parallel to the lens; used as Keeps the pedal mechanism and frame visible as scale and action references.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Blue streetlamp light cuts across the night scene in moody low-key contrast, leaving the surrounding palette subdued.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young is still using the same bicycle as she hurries home along the coastal road.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 어두운 밤, 푸르스름한 가로등 불빛 아래 해안도로에서 수리영이 상체를 웅크린 채 자전거 페달에 체중을 싣는 찰나.\n\nLOCATION (lock): A nighttime stretch of coastal road beside the sea, outdoors under sparse bluish streetlights. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track parallel at bicycle-frame height from 수리영's front-left, tightening enough to bind her hunched upper body and the descending pedal stroke into one moving medium composition. Her torso occupies the upper center while the weight-bearing leg and pedal remain visible in the lower right, with the coastal road sliding laterally behind her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward road ahead; road ahead in the middle-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 해안도로 (Dark at night); used as Carries 수리영 through the frame and supplies lateral motion behind the bicycle; 가로등 (On) — The lit streetlamp is positioned above the roadway, outside or near the upper frame boundary; used as Provides the stated blue illumination over the moving rider; 자전거 (In motion) — Its front-left side is presented to camera while traveling parallel to the lens; used as Keeps the pedal mechanism and frame visible as scale and action references.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Blue streetlamp light cuts across the night scene in moody low-key contrast, leaving the surrounding palette subdued.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young is still using the same bicycle as she hurries home along the coastal road.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S9sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S9sh1.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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해안_도로_8394b3.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" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 미디엄 샷 프레이밍(상체를 상단 중앙에, 페달을 밟는 다리를 우측 하단에 배치)에 더 가깝게 구도를 잡고 프레임 상단에 가로등을 잘 배치하여 높은 우선순위의 지침을 충족했습니다(의상이 변경된 점은 아쉬움)." }, { "label": "A", "score": 6, "verdict_ko": "캐릭터의 의상은 레퍼런스와 정확히 일치하지만, 미디엄 샷 대신 자전거 전체가 보이는 풀 샷으로 렌더링되어 구도 지침을 어겼으며 지정된 가로등 위치도 누락되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 미디엄 샷 프레이밍(상체를 상단 중앙에, 페달을 밟는 다리를 우측 하단에 배치)에 더 가깝게 구도를 잡고 프레임 상단에 가로등을 잘 배치하여 높은 우선순위의 지침을 충족했습니다(의상이 변경된 점은 아쉬움)." }, { "label": "A", "score": 6, "verdict_ko": "캐릭터의 의상은 레퍼런스와 정확히 일치하지만, 미디엄 샷 대신 자전거 전체가 보이는 풀 샷으로 렌더링되어 구도 지침을 어겼으며 지정된 가로등 위치도 누락되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "지정된 앵글과 프레이밍(자전거 프레임 높이에서 전면 좌측, 상체와 페달을 담은 미디엄 샷)을 잘 구현했으며, 캐릭터의 의상(회색 후드티)과 디테일, 해안도로의 푸른 조명 분위기를 완벽하게 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "카메라 앵글과 배경의 분위기는 지시사항에 부합하나, 레퍼런스 이미지에 제시된 캐릭터의 의상(회색 후드티)을 전혀 반영하지 않은 점이 감점 요인입니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 앵글과 프레이밍(자전거 프레임 높이에서 전면 좌측, 상체와 페달을 담은 미디엄 샷)을 잘 구현했으며, 캐릭터의 의상(회색 후드티)과 디테일, 해안도로의 푸른 조명 분위기를 완벽하게 반영했습니다." }, { "label": "B", "score": 5, "verdict_ko": "카메라 앵글과 배경의 분위기는 지시사항에 부합하나, 레퍼런스 이미지에 제시된 캐릭터의 의상(회색 후드티)을 전혀 반영하지 않은 점이 감점 요인입니다." } ] }, "combined": { "totals": { "A": 15, "B": 13 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 15, "B": 13 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 미디엄 샷 프레이밍(상체를 상단 중앙에, 페달을 밟는 다리를 우측 하단에 배치)에 더 가깝게 구도를 잡고 프레임 상단에 가로등을 잘 배치하여 높은 우선순위의 지침을 충족했습니다(의상이 변경된 점은 아쉬움)." }, { "label": "A", "score": 6, "verdict_ko": "캐릭터의 의상은 레퍼런스와 정확히 일치하지만, 미디엄 샷 대신 자전거 전체가 보이는 풀 샷으로 렌더링되어 구도 지침을 어겼으며 지정된 가로등 위치도 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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해안_도로_8394b3.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "카메라가 지정된 것보다 멀리 위치하여, 인물의 상체와 페달 밟는 다리를 타이트하게 잡는 미디엄 샷이 아닌 자전거 전체가 보이는 풀 샷이 되었습니다.", "fix_en": "The shot is framed as a full shot showing the entire bicycle instead of a medium composition bounding the upper body and descending pedal stroke; unfixable without moving the camera closer.", "unfixable": true }, { "issue_ko": "원본 배경 이미지에는 없던 커다란 바위들이 왼쪽 바다 영역에 임의로 추가되었습니다.", "fix_en": "Remove the large rocks from the water on the left side, restoring the empty dark sea from the original background reference.", "unfixable": false }, { "issue_ko": "앞바퀴와 차체 사이의 허공(도로 위)에 흰색 운동화를 신은 정체불명의 발이 하나 더 떠 있습니다.", "fix_en": "Remove the extra floating white shoe and foot from behind the front wheel, restoring the dark road surface behind it.", "unfixable": false }, { "issue_ko": "자전거의 페달 중심축(크랭크)에서 위쪽 허공으로 뻗어 나가는 불필요한 금속 막대가 잘못 생성되어 있습니다.", "fix_en": "Remove the extra metal crank arm pointing upwards from the bicycle's center bracket.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Rotate the crank and restage the weight-bearing leg so the descending foot and pedal appear in the lower-right of the frame while preserving the rider, bicycle, and camera placement.", "issue_ko": "체중이 실린 하강 페달과 다리가 화면 오른쪽 아래가 아니라 아래쪽 중앙에 배치되어 있다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the large rocks from the water on the left side, restoring the empty dark sea from the original background reference.\n- Remove the extra floating white shoe and foot from behind the front wheel, restoring the dark road surface behind it.\n- Remove the extra metal crank arm pointing upwards from the bicycle's center bracket.\n- Rotate the crank and restage the weight-bearing leg so the descending foot and pedal appear in the lower-right of the frame while preserving the rider, bicycle, and camera placement.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 5, "verdict_ko": "지정된 원본 배경과 인물의 외형, 자세를 정확히 유지했으나, 프레임이 지시된 미디엄 샷보다 넓게 잡혀 아쉽습니다." }, { "label": "A", "score": 3, "verdict_ko": "원본 배경의 바다에 없던 거대한 바위를 임의로 추가하는 치명적인 오류를 범했으며, 미디엄 샷 지시도 따르지 않았습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 5, "verdict_ko": "지정된 원본 배경과 인물의 외형, 자세를 정확히 유지했으나, 프레임이 지시된 미디엄 샷보다 넓게 잡혀 아쉽습니다." }, { "label": "A", "score": 3, "verdict_ko": "원본 배경의 바다에 없던 거대한 바위를 임의로 추가하는 치명적인 오류를 범했으며, 미디엄 샷 지시도 따르지 않았습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 원본 배경을 그대로 유지하면서 인물의 의상, 포즈, 자전거 구도를 훌륭하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "원본 배경의 바다 영역에 없던 거대한 바위들을 임의로 추가하여 배경을 정확히 유지하라는 핵심 지시를 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 원본 배경을 그대로 유지하면서 인물의 의상, 포즈, 자전거 구도를 훌륭하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "원본 배경의 바다 영역에 없던 거대한 바위들을 임의로 추가하여 배경을 정확히 유지하라는 핵심 지시를 위반했습니다." } ] }, "combined": { "totals": { "A": 6, "B": 13 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S9sh1__bgfirst_bg.png", "bg_asset_id": "31c7a969-a098-46d3-a284-6704e6eb6834", "bg_record_key": "S9sh1::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "해안 도로", "groupbg_asset_id": "8f0394d6-f50d-4f43-baa3-d2420dc9b871" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S10sh2::bgfirst_bg": { "input_fingerprint": "0936f64df8412a17", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 현관 철문 틈새에 끼워져 있던 광고 전단지를 꽉 쥔 남성의 손 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop home's metal entrance door at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in at hand height beside the doorway until the intruder's gripping hand occupies most of the insert, while the advertising leaflet remains below the nonhuman-object limit and the narrow door gap survives along one edge. Crop out the man's face and body, using the iron door edge as a rigid vertical against the compressed fingers and partly folded leaflet.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 광고 전단지 (Pulled from the gap and tightly gripped) — Part of its printed advertising face is visible to camera, though the grip and fold obscure the specific content; used as The key manipulated object compressed inside the intruder's fist beside the gap; 현관 철문과 문틈 (Closed and still locked at this instant) — The door edge is nearly side-on to camera, exposing the narrow gap beside the hand; used as A narrow vertical boundary framing the hand and the intended insertion point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Nighttime ambient light is kept sparse and low-key, emphasizing the hand, leaflet, and door gap without inventing a specific source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 현관 철문 틈새에 끼워져 있던 광고 전단지를 꽉 쥔 남성의 손 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop home's metal entrance door at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in at hand height beside the doorway until the intruder's gripping hand occupies most of the insert, while the advertising leaflet remains below the nonhuman-object limit and the narrow door gap survives along one edge. Crop out the man's face and body, using the iron door edge as a rigid vertical against the compressed fingers and partly folded leaflet.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 광고 전단지 (Pulled from the gap and tightly gripped) — Part of its printed advertising face is visible to camera, though the grip and fold obscure the specific content; used as The key manipulated object compressed inside the intruder's fist beside the gap; 현관 철문과 문틈 (Closed and still locked at this instant) — The door edge is nearly side-on to camera, exposing the narrow gap beside the hand; used as A narrow vertical boundary framing the hand and the intended insertion point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Nighttime ambient light is kept sparse and low-key, emphasizing the hand, leaflet, and door gap without inventing a specific source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S10sh2__bgfirst_bg.png", "asset_id": "eb715eee-cd6a-4642-8faa-dc444a6fef4f", "input_asset_ids": [ "d0029348-fa7d-4e1b-b564-94fd7c3745f9", "a589f40e-c1dd-43f1-b838-b4890c1f9762", "605fee83-c4e3-45fc-8b9a-bd7d7a504772" ] }, "S10sh2": { "input_fingerprint": "38fabe626ac18ca7", "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): 현관 철문 틈새에 끼워져 있던 광고 전단지를 꽉 쥔 남성의 손 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop home's metal entrance door at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in at hand height beside the doorway until the intruder's gripping hand occupies most of the insert, while the advertising leaflet remains below the nonhuman-object limit and the narrow door gap survives along one edge. Crop out the man's face and body, using the iron door edge as a rigid vertical against the compressed fingers and partly folded leaflet.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 광고 전단지 (Pulled from the gap and tightly gripped) — Part of its printed advertising face is visible to camera, though the grip and fold obscure the specific content; used as The key manipulated object compressed inside the intruder's fist beside the gap; 현관 철문과 문틈 (Closed and still locked at this instant) — The door edge is nearly side-on to camera, exposing the narrow gap beside the hand; used as A narrow vertical boundary framing the hand and the intended insertion point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Nighttime ambient light is kept sparse and low-key, emphasizing the hand, leaflet, and door gap without inventing a specific source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The advertisement flyer has been pulled from the iron door gap and is now held in the stranger’s hand for use on the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 침입 남성 (한국인 성인 남성, 가려진 얼굴, 짙은색 머리카락, 불분명한 머리 윤곽) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 현관 철문 틈새에 끼워져 있던 광고 전단지를 꽉 쥔 남성의 손 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop home's metal entrance door at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in at hand height beside the doorway until the intruder's gripping hand occupies most of the insert, while the advertising leaflet remains below the nonhuman-object limit and the narrow door gap survives along one edge. Crop out the man's face and body, using the iron door edge as a rigid vertical against the compressed fingers and partly folded leaflet.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 광고 전단지 (Pulled from the gap and tightly gripped) — Part of its printed advertising face is visible to camera, though the grip and fold obscure the specific content; used as The key manipulated object compressed inside the intruder's fist beside the gap; 현관 철문과 문틈 (Closed and still locked at this instant) — The door edge is nearly side-on to camera, exposing the narrow gap beside the hand; used as A narrow vertical boundary framing the hand and the intended insertion point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Nighttime ambient light is kept sparse and low-key, emphasizing the hand, leaflet, and door gap without inventing a specific source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The advertisement flyer has been pulled from the iron door gap and is now held in the stranger’s hand for use on the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 침입 남성 (한국인 성인 남성, 가려진 얼굴, 짙은색 머리카락, 불분명한 머리 윤곽) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 현관 철문 틈새에 끼워져 있던 광고 전단지를 꽉 쥔 남성의 손 클로즈업.\n\nLOCATION (lock): On the rooftop landing directly outside the rooftop home's metal entrance door at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nSTRUCTURE LOOK AUTHORITY: the attached STRUCTURE LOOK photograph is the identity of the fixed structure at this location — wherever that structure appears in the frame, its shape, proportions, openings, materials and colors are LOCKED to it. The LOCATION PHOTOGRAPH remains the authority for this shot's sub-space, surroundings, time of day and lighting. If the two conflict on the structure itself, the STRUCTURE LOOK photo wins; for everything else, the LOCATION PHOTOGRAPH wins.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Dolly in at hand height beside the doorway until the intruder's gripping hand occupies most of the insert, while the advertising leaflet remains below the nonhuman-object limit and the narrow door gap survives along one edge. Crop out the man's face and body, using the iron door edge as a rigid vertical against the compressed fingers and partly folded leaflet.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 광고 전단지 (Pulled from the gap and tightly gripped) — Part of its printed advertising face is visible to camera, though the grip and fold obscure the specific content; used as The key manipulated object compressed inside the intruder's fist beside the gap; 현관 철문과 문틈 (Closed and still locked at this instant) — The door edge is nearly side-on to camera, exposing the narrow gap beside the hand; used as A narrow vertical boundary framing the hand and the intended insertion point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Nighttime ambient light is kept sparse and low-key, emphasizing the hand, leaflet, and door gap without inventing a specific source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The advertisement flyer has been pulled from the iron door gap and is now held in the stranger’s hand for use on the lock.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 옥탑방 침입 남성 (한국인 성인 남성, 가려진 얼굴, 짙은색 머리카락, 불분명한 머리 윤곽) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S10sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S10sh2.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 침입 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L03B02.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/background_chain/seed_bg_rooftop_residence_sel.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 침입 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "철문 모서리를 수직 기준으로 삼아 문틈과 전단지를 꽉 쥔 손을 명시된 인서트 클로즈업 구도로 정확하게 연출했습니다." }, { "label": "A", "score": 4, "verdict_ko": "문의 정면만 나타날 뿐, 지시된 철문 모서리와 좁은 문틈이 프레임에 포함되지 않아 구도 요구사항을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "철문 모서리를 수직 기준으로 삼아 문틈과 전단지를 꽉 쥔 손을 명시된 인서트 클로즈업 구도로 정확하게 연출했습니다." }, { "label": "A", "score": 4, "verdict_ko": "문의 정면만 나타날 뿐, 지시된 철문 모서리와 좁은 문틈이 프레임에 포함되지 않아 구도 요구사항을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "명시된 측면 앵글을 통해 철문 가장자리를 수직선으로 활용하고 문틈 앞 전단지를 쥔 손을 프레이밍에 정확히 담아냈습니다." }, { "label": "B", "score": 4, "verdict_ko": "카메라가 문의 정면을 향하고 있어, 지시사항의 핵심인 '문 가장자리가 측면으로 보이는 구도'와 '문틈' 묘사에 실패했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "명시된 측면 앵글을 통해 철문 가장자리를 수직선으로 활용하고 문틈 앞 전단지를 쥔 손을 프레이밍에 정확히 담아냈습니다." }, { "label": "A", "score": 4, "verdict_ko": "카메라가 문의 정면을 향하고 있어, 지시사항의 핵심인 '문 가장자리가 측면으로 보이는 구도'와 '문틈' 묘사에 실패했습니다." } ] }, "combined": { "totals": { "A": 8, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 8, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "철문 모서리를 수직 기준으로 삼아 문틈과 전단지를 꽉 쥔 손을 명시된 인서트 클로즈업 구도로 정확하게 연출했습니다." }, { "label": "A", "score": 4, "verdict_ko": "문의 정면만 나타날 뿐, 지시된 철문 모서리와 좁은 문틈이 프레임에 포함되지 않아 구도 요구사항을 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L03B02.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/background_chain/seed_bg_rooftop_residence_sel.png" }, { "label": "CHARACTER REFERENCE — 옥탑방 침입 남성: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프롬프트에 '텍스트 없음'이 명시되어 있으나, 전단지에 한글 텍스트('광고 전단지', '열쇠' 등)가 선명하게 생성되었습니다.", "fix_en": "Replace all legible text and letters on the folded leaflet with abstract, illegible colored graphics and shapes that suggest printed advertising without forming actual words.", "unfixable": false }, { "issue_ko": "프롬프트에서 문은 '닫혀 있고 잠긴 상태(Closed and still locked)'여야 하지만, 이미지에서는 도어 프레임의 래치 구멍(스트라이크 플레이트)이 드러난 채 문이 살짝 열려 있습니다.", "fix_en": "Adjust the door and frame so the door is tightly closed and flush against the frame, completely concealing the empty strike plate holes.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Crop tighter around the gripping fist, folded leaflet, and door edge so the hand dominates the insert and the vertical gap sits near the left frame edge, removing excess door panel and sleeve area.", "issue_ko": "프레이밍이 지시보다 느슨해 손이 화면 대부분을 차지하지 못하고, 문틈도 프레임 한쪽 가장자리가 아니라 화면 안쪽에 놓여 있다.", "unfixable": false }, { "fix_en": "Reduce the black opening between the door and jamb to a very thin vertical sliver so the door reads as closed.", "issue_ko": "철문과 문틀 사이의 검은 간격이 너무 넓어 잠긴 채 닫힌 문의 좁은 틈이 아니라 문이 열린 상태처럼 보인다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace all legible text and letters on the folded leaflet with abstract, illegible colored graphics and shapes that suggest printed advertising without forming actual words.\n- Adjust the door and frame so the door is tightly closed and flush against the frame, completely concealing the empty strike plate holes.\n- Crop tighter around the gripping fist, folded leaflet, and door edge so the hand dominates the insert and the vertical gap sits near the left frame edge, removing excess door panel and sleeve area.\n- Reduce the black opening between the door and jamb to a very thin vertical sliver so the door reads as closed.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 클로즈업 구도와 전단지를 쥔 손의 형태를 정확히 구현했으며, 철문 단면의 물리적 구조가 자연스럽습니다." }, { "label": "A", "score": 5, "verdict_ko": "구도와 피사체 묘사는 우수하나, 문 손잡이와 떨어진 문짝 단면에 비정상적인 구조물이 생성된 물리적 오류가 있습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 클로즈업 구도와 전단지를 쥔 손의 형태를 정확히 구현했으며, 철문 단면의 물리적 구조가 자연스럽습니다." }, { "label": "A", "score": 5, "verdict_ko": "구도와 피사체 묘사는 우수하나, 문 손잡이와 떨어진 문짝 단면에 비정상적인 구조물이 생성된 물리적 오류가 있습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 프레이밍과 피사체(손, 전단지)를 잘 구현했으며, 프롬프트에 명시된 '닫혀 있는 상태에서 드러난 좁은 문틈'을 명확히 묘사하여 지시사항에 충실합니다." }, { "label": "A", "score": 4, "verdict_ko": "대부분의 프레이밍과 요소는 일치하나, 프롬프트에서 요구한 '드러난 좁은 문틈(gap)'이 시각적으로 명확하게 표현되지 않았습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 프레이밍과 피사체(손, 전단지)를 잘 구현했으며, 프롬프트에 명시된 '닫혀 있는 상태에서 드러난 좁은 문틈'을 명확히 묘사하여 지시사항에 충실합니다." }, { "label": "B", "score": 4, "verdict_ko": "대부분의 프레이밍과 요소는 일치하나, 프롬프트에서 요구한 '드러난 좁은 문틈(gap)'이 시각적으로 명확하게 표현되지 않았습니다." } ] }, "combined": { "totals": { "A": 12, "B": 11 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S10sh2__bgfirst_bg.png", "bg_asset_id": "eb715eee-cd6a-4642-8faa-dc444a6fef4f", "bg_record_key": "S10sh2::bgfirst_bg", "chain_winner": false, "authority": "plate", "seed_attached": true }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready", "plate_select": { "candidates": { "A": "L03B01", "B": "L03B02" }, "assigned": "L03B02", "choice": "Candidate B", "confident": true, "reason_ko": "지문에서 '현관 철문 틈새'를 언급하고 있으므로 금속 재질의 출입문이 있는 공간이 필요합니다. 두 후보 모두 철문을 포함하고 있으나, 현재 할당된 후보 B가 해당 공간을 명확히 보여주므로 이를 유지합니다.", "kept": "L03B02" } }, "S10sh4::variants": { "author_fp": "daccace7ddec624b", "author": { "variants": [ { "approach_ko": "창문을 정면의 엄격한 틀로 삼아 두 그림자의 공격 동작을 또렷하고 평면적인 실루엣 구도로 압축한다.", "prompt_en": "Use a restrained, near-normal perspective and square-on architectural composition, keeping the entire illuminated pane and its fixed frame visually disciplined. Expose for the interior light so the overlapping human shadows become crisp, opaque contours: the attacker’s open-mouthed profile dominates above while the victim’s lower recoil remains clearly legible. Keep exterior surfaces subdued and the glass physically present, making the window feel like a hard boundary that withholds the assault." }, { "approach_ko": "가까운 광각의 미세한 원근감과 거친 창틀의 존재감으로 관객이 창밖에 붙어 엿보는 듯한 압박을 만든다.", "prompt_en": "Photograph from close to the window with a moderately wide perspective while preserving the locked medium framing and nearly straight-on view. Let the window frame carry stronger edge presence and subtle perspective tension around the luminous pane, while focus remains on the shadow geometry across the glass. Hold the interior source stark but allow a narrow, natural penumbra along the moving silhouettes, making the attacker’s enlarged upper form feel urgently close to the victim’s shoulder without revealing either person directly." } ] }, "reused": true }, "S10sh4": { "input_fingerprint": "682cfcd0c40eb25e", "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): 옥탑방의 닫힌 창문 불빛 위로, 누군가 다른 사람의 어깨 쪽을 향해 입을 벌려 덮치는 듯한 거대한 형상이 그림자로만 비친 모습.\n\nLOCATION (lock): On the rooftop outside the illuminated window of the rooftop home, looking at the attack only as shadows reflected in the glass. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold static at chest height directly before the closed window, framing its lit area as a contained medium tableau rather than entering the room. The 그림자 속 공격자 overlaps the upper center in profile with mouth open toward the 그림자 속 피해자's shoulder, while the victim's recoiling silhouette remains lower and slightly to the side so the attack reads only through shadow geometry.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 그림자 속 공격자 silhouette in the upper-center of the frame, background, moves toward victim's shoulder; 그림자 속 피해자 silhouette in the middle-right of the frame, background; closed illuminated window in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 닫힌 창문 (Closed and illuminated) — The camera faces the closed pane nearly straight on; illumination from inside reveals only the two overlapping shadows, not clear room detail; used as Contains both overlapping silhouettes and withholds direct sight of the assault; 창문틀 (Fixed around the closed window) — Its front side faces the camera nearly squarely around the lit pane; used as Forms the outer boundary that traps the attack within the visible window area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior illumination behind the closed window renders the assault as stark dark silhouettes with tense low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room’s iron entrance door has been unlocked and left slightly open by the intruder as he attacks Min-sook inside.\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.\n\nSHOT TEXT (authoritative, Korean): 옥탑방의 닫힌 창문 불빛 위로, 누군가 다른 사람의 어깨 쪽을 향해 입을 벌려 덮치는 듯한 거대한 형상이 그림자로만 비친 모습.\n\nLOCATION (lock): On the rooftop outside the illuminated window of the rooftop home, looking at the attack only as shadows reflected in the glass. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold static at chest height directly before the closed window, framing its lit area as a contained medium tableau rather than entering the room. The 그림자 속 공격자 overlaps the upper center in profile with mouth open toward the 그림자 속 피해자's shoulder, while the victim's recoiling silhouette remains lower and slightly to the side so the attack reads only through shadow geometry.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 그림자 속 공격자 silhouette in the upper-center of the frame, background, moves toward victim's shoulder; 그림자 속 피해자 silhouette in the middle-right of the frame, background; closed illuminated window in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 닫힌 창문 (Closed and illuminated) — The camera faces the closed pane nearly straight on; illumination from inside reveals only the two overlapping shadows, not clear room detail; used as Contains both overlapping silhouettes and withholds direct sight of the assault; 창문틀 (Fixed around the closed window) — Its front side faces the camera nearly squarely around the lit pane; used as Forms the outer boundary that traps the attack within the visible window area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior illumination behind the closed window renders the assault as stark dark silhouettes with tense low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room’s iron entrance door has been unlocked and left slightly open by the intruder as he attacks Min-sook inside.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, near-normal perspective and square-on architectural composition, keeping the entire illuminated pane and its fixed frame visually disciplined. Expose for the interior light so the overlapping human shadows become crisp, opaque contours: the attacker’s open-mouthed profile dominates above while the victim’s lower recoil remains clearly legible. Keep exterior surfaces subdued and the glass physically present, making the window feel like a hard boundary that withholds the assault.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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): 옥탑방의 닫힌 창문 불빛 위로, 누군가 다른 사람의 어깨 쪽을 향해 입을 벌려 덮치는 듯한 거대한 형상이 그림자로만 비친 모습.\n\nLOCATION (lock): On the rooftop outside the illuminated window of the rooftop home, looking at the attack only as shadows reflected in the glass. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold static at chest height directly before the closed window, framing its lit area as a contained medium tableau rather than entering the room. The 그림자 속 공격자 overlaps the upper center in profile with mouth open toward the 그림자 속 피해자's shoulder, while the victim's recoiling silhouette remains lower and slightly to the side so the attack reads only through shadow geometry.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 그림자 속 공격자 silhouette in the upper-center of the frame, background, moves toward victim's shoulder; 그림자 속 피해자 silhouette in the middle-right of the frame, background; closed illuminated window in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 닫힌 창문 (Closed and illuminated) — The camera faces the closed pane nearly straight on; illumination from inside reveals only the two overlapping shadows, not clear room detail; used as Contains both overlapping silhouettes and withholds direct sight of the assault; 창문틀 (Fixed around the closed window) — Its front side faces the camera nearly squarely around the lit pane; used as Forms the outer boundary that traps the attack within the visible window area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior illumination behind the closed window renders the assault as stark dark silhouettes with tense low-key contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room’s iron entrance door has been unlocked and left slightly open by the intruder as he attacks Min-sook inside.\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 window with a moderately wide perspective while preserving the locked medium framing and nearly straight-on view. Let the window frame carry stronger edge presence and subtle perspective tension around the luminous pane, while focus remains on the shadow geometry across the glass. Hold the interior source stark but allow a narrow, natural penumbra along the moving silhouettes, making the attacker’s enlarged upper form feel urgently close to the victim’s shoulder without revealing either person directly.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": 9, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지시된 구도와 창문 너머의 그림자 실루엣을 정확히 연출했으며, 레퍼런스에 제시된 철문과 벽면의 물리적 구조를 자연스럽게 반영했습니다." }, { "label": "B", "score": 3, "verdict_ko": "창문이 옥상 한가운데 독립된 상자처럼 묘사되고 레퍼런스의 철문이 벽면 없이 허공에 떠 있는 등 공간의 건축적 구조를 심각하게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S10sh2_sel.png" } ], "critique": { "issues": [ { "issue_ko": "왼쪽에 보이는 철문이 너무 활짝 열려 있습니다. 지시문(carried state)에 따라 아주 살짝만 틈이 벌어진(slightly open) 상태여야 합니다.", "fix_en": "Close the door on the far left so it is only slightly ajar, leaving just a narrow crack instead of being swung wide open.", "unfixable": false }, { "issue_ko": "실내에서 유리창에 비친 그림자가 창문의 중앙 수직 프레임(창틀)을 가리고 그 위에 덧씌워져 있습니다. 그림자는 유리 뒤에 있어야 합니다.", "fix_en": "Restore the center vertical window frame so it is physically in front of the shadows, confining the dark silhouettes strictly to the glass panes.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Close the door on the far left so it is only slightly ajar, leaving just a narrow crack instead of being swung wide open.\n- Restore the center vertical window frame so it is physically in front of the shadows, confining the dark silhouettes strictly to the glass panes.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "창문 너머의 그림자 실루엣과 약간 열린 철문 등 프롬프트의 요구사항을 훌륭하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "왼쪽 열린 문 부분에 손잡이가 두 개 생성되는 등 구조적인 오류(Hard Violation)가 발생했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "창문 너머의 그림자 실루엣과 약간 열린 철문 등 프롬프트의 요구사항을 훌륭하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "왼쪽 열린 문 부분에 손잡이가 두 개 생성되는 등 구조적인 오류(Hard Violation)가 발생했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "창문에 비친 그림자의 디테일과 살짝 열린 철문의 상태를 프롬프트에 맞게 매우 잘 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "프롬프트의 전반적인 요구사항을 잘 충족했으나, 철문 손잡이와 창문 잠금장치 등 미세한 디테일의 완성도가 B에 비해 약간 떨어집니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "창문에 비친 그림자의 디테일과 살짝 열린 철문의 상태를 프롬프트에 맞게 매우 잘 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "프롬프트의 전반적인 요구사항을 잘 충족했으나, 철문 손잡이와 창문 잠금장치 등 미세한 디테일의 완성도가 B에 비해 약간 떨어집니다." } ] }, "combined": { "totals": { "A": 15, "B": 10 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S10sh2" }, "lane_policy": "ab_select_bypass:bg_only:share_plan_prev_bgonly", "plate_select": { "candidates": { "A": "L03B01", "B": "L03B02" }, "assigned": "L03B02", "choice": "Candidate B", "confident": true, "reason_ko": "\"닫힌 창문 불빛 위로... 그림자로만 비친 모습\"이라는 묘사에 부합하도록, 블라인드가 닫혀 내부 실루엣(그림자)이 비치기 적합한 창문 구조를 가진 Candidate B가 적절합니다.", "kept": "L03B02" } }, "S11sh3::variants": { "author_fp": "f37da17636a0918d", "author": { "variants": [ { "approach_ko": "얕은 심도와 절제된 압축감으로 어둠 속 반쯤 열린 철문에 시선을 고정하는 긴장감 있는 접근.", "prompt_en": "Use a restrained, slightly compressed lens feel with shallow depth, letting the lower-left foreground and near threshold fall softly out of focus while the half-open metal door resolves as the sharp visual endpoint. Hold the camera at arriving eye height and preserve a measured field of darkness around the doorway; faint ambient light should skim the worn metal face and barely separate the narrow opening behind its edge, creating quiet, watchful tension." }, { "approach_ko": "문턱에서 철문까지의 거리와 빈 옥상 공간을 깊게 살려 버려진 입구의 불길한 정적을 강조하는 접근.", "prompt_en": "Favor a more spatial, deep-focus treatment that makes the foreground threshold a firm compositional boundary and clearly carries the eye across the empty rooftop depth to the half-open metal door. Keep the oblique door face and the darkness beyond its edge simultaneously legible, using subdued ambient falloff and minimal surface glints rather than a spotlight; emphasize physical distance, stillness, and the unsettling vacancy surrounding the unsecured entrance." } ] }, "reused": false }, "S11sh3": { "input_fingerprint": "7d588042f60d276b", "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): 옥상에 다다른 수리영의 시선 끝, 짙은 어둠 속에서 문틈이 반쯤 열린 채 방치된 옥탑방 현관 철문의 모습.\n\nLOCATION (lock): At the top of the exterior rooftop stairs, facing the rooftop home's half-open metal entrance door in the darkness. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Continue the dolly-in from 수리영's eye height just behind her shoulder, tightening past her blurred shoulder so the half-open iron entrance door becomes the endpoint across the rooftop. Keep the door below excessive object occupancy and retain enough surrounding threshold space to make its abandoned opening physically legible.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, foreground, looks toward half-open entrance door; half-open entrance door in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 현관 철문 (Half-open) — The door is seen obliquely with its front face and the opening behind its edge both visible; used as The focal endpoint of 수리영's sightline across the rooftop; 옥상 문턱 (Open passage onto the rooftop) — Its inner edge crosses the lower foreground before the rooftop opens toward the entrance; used as Foreground boundary that locates 수리영 at the point of arrival.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Night darkness is preserved in subdued low-key contrast, with only enough ambient visibility to define 수리영's shoulder and the half-open door.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The iron entrance door remains unsecured and half open after the stranger’s forced entry.\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.\n\nSHOT TEXT (authoritative, Korean): 옥상에 다다른 수리영의 시선 끝, 짙은 어둠 속에서 문틈이 반쯤 열린 채 방치된 옥탑방 현관 철문의 모습.\n\nLOCATION (lock): At the top of the exterior rooftop stairs, facing the rooftop home's half-open metal entrance door in the darkness. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Continue the dolly-in from 수리영's eye height just behind her shoulder, tightening past her blurred shoulder so the half-open iron entrance door becomes the endpoint across the rooftop. Keep the door below excessive object occupancy and retain enough surrounding threshold space to make its abandoned opening physically legible.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, foreground, looks toward half-open entrance door; half-open entrance door in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 현관 철문 (Half-open) — The door is seen obliquely with its front face and the opening behind its edge both visible; used as The focal endpoint of 수리영's sightline across the rooftop; 옥상 문턱 (Open passage onto the rooftop) — Its inner edge crosses the lower foreground before the rooftop opens toward the entrance; used as Foreground boundary that locates 수리영 at the point of arrival.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Night darkness is preserved in subdued low-key contrast, with only enough ambient visibility to define 수리영's shoulder and the half-open door.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The iron entrance door remains unsecured and half open after the stranger’s forced entry.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, slightly compressed lens feel with shallow depth, letting the lower-left foreground and near threshold fall softly out of focus while the half-open metal door resolves as the sharp visual endpoint. Hold the camera at arriving eye height and preserve a measured field of darkness around the doorway; faint ambient light should skim the worn metal face and barely separate the narrow opening behind its edge, 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): night.\n\nSHOT TEXT (authoritative, Korean): 옥상에 다다른 수리영의 시선 끝, 짙은 어둠 속에서 문틈이 반쯤 열린 채 방치된 옥탑방 현관 철문의 모습.\n\nLOCATION (lock): At the top of the exterior rooftop stairs, facing the rooftop home's half-open metal entrance door in the darkness. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Continue the dolly-in from 수리영's eye height just behind her shoulder, tightening past her blurred shoulder so the half-open iron entrance door becomes the endpoint across the rooftop. Keep the door below excessive object occupancy and retain enough surrounding threshold space to make its abandoned opening physically legible.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, foreground, looks toward half-open entrance door; half-open entrance door in the middle-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 현관 철문 (Half-open) — The door is seen obliquely with its front face and the opening behind its edge both visible; used as The focal endpoint of 수리영's sightline across the rooftop; 옥상 문턱 (Open passage onto the rooftop) — Its inner edge crosses the lower foreground before the rooftop opens toward the entrance; used as Foreground boundary that locates 수리영 at the point of arrival.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Night darkness is preserved in subdued low-key contrast, with only enough ambient visibility to define 수리영's shoulder and the half-open door.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The iron entrance door remains unsecured and half open after the stranger’s forced entry.\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):\nFavor a more spatial, deep-focus treatment that makes the foreground threshold a firm compositional boundary and clearly carries the eye across the empty rooftop depth to the half-open metal door. Keep the oblique door face and the darkness beyond its edge simultaneously legible, using subdued ambient falloff and minimal surface glints rather than a spotlight; emphasize physical distance, stillness, and the unsettling vacancy surrounding the unsecured entrance.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": "옥탑방 전면에 위치해야 할 창문이 누락되어 구조물(Structure Look) 기준을 위반했습니다." }, { "label": "B", "score": 7, "verdict_ko": "지정된 숄더샷 구도와 야간 조명을 정확히 연출했으며, 옥탑방 전면 창문 등 기준 이미지의 공간 디테일을 충실히 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L11B04.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/79670d0c-2096-4d40-88cf-30b407411548/images/background_chain/seed_bg_rooftop_residence_sel.png" } ], "critique": { "issues": [ { "issue_ko": "문의 형태와 재질이 구조물 레퍼런스(매끄러운 진회색)가 아닌 장소 레퍼런스(녹슬고 도어 클로저가 부착됨)를 따르고 있습니다.", "fix_en": "Change the door's surface to a smooth, plain dark-grey metal and remove the door closer mechanism at the top, matching the structure look reference.", "unfixable": false }, { "issue_ko": "구조물 레퍼런스의 우측 지붕 위에 있는 커다란 은색 원통형 물탱크가 이미지에서 누락되었습니다.", "fix_en": "Add the large, cylindrical silver water tank to the right side of the rooftop structure's flat roof, matching the structure look reference.", "unfixable": false }, { "issue_ko": "조명이 은은한 앰비언트 라이트여야 한다는 지시와 달리, 문과 주변 벽에 강한 스포트라이트가 비추고 있습니다.", "fix_en": "Darken the brightly lit area on the door and surrounding wall to remove the spotlight effect, creating a uniform, subdued ambient darkness.", "unfixable": false }, { "issue_ko": "전경 하단을 가로질러 도착 지점을 나타내야 하는 옥상 문턱(계단과 옥상의 경계)이 묘사되지 않고 초록색 바닥만 이어져 있습니다.", "fix_en": "Add a horizontal concrete step edge across the very bottom of the frame, right below and behind the character, to represent the entry threshold to the rooftop.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Remove the entire foreground person and reconstruct the occluded rooftop view while preserving the existing camera position and perspective.", "issue_ko": "인물이 없어야 하는 숏인데 화면 왼쪽 전경을 인물의 머리와 상반신이 크게 점유하고 있다.", "unfixable": false }, { "fix_en": "Reveal the existing rooftop-arrival threshold edge across the lower foreground without changing the camera position or rooftop perspective.", "issue_ko": "옥상 도착 지점을 규정해야 할 문턱의 안쪽 경계가 하단 전경을 가로지르지 않아 전경 경계가 명확하지 않다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the door's surface to a smooth, plain dark-grey metal and remove the door closer mechanism at the top, matching the structure look reference.\n- Add the large, cylindrical silver water tank to the right side of the rooftop structure's flat roof, matching the structure look reference.\n- Darken the brightly lit area on the door and surrounding wall to remove the spotlight effect, creating a uniform, subdued ambient darkness.\n- Add a horizontal concrete step edge across the very bottom of the frame, right below and behind the character, to represent the entry threshold to the rooftop.\n- Remove the entire foreground person and reconstruct the occluded rooftop view while preserving the existing camera position and perspective.\n- Reveal the existing rooftop-arrival threshold edge across the lower foreground without changing the camera position or rooftop perspective.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 좌측 하단 인물의 어깨 너머 앵글을 정확히 구현하였으며, 야간의 조명과 옥상 구조를 레퍼런스에 맞게 충실히 반영했습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레임 레이아웃에서 명시적으로 요구한 전경의 인물을 완전히 누락하였으며, 레퍼런스와 달리 물탱크가 옥탑 지붕 위에 잘못 배치되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 좌측 하단 인물의 어깨 너머 앵글을 정확히 구현하였으며, 야간의 조명과 옥상 구조를 레퍼런스에 맞게 충실히 반영했습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레임 레이아웃에서 명시적으로 요구한 전경의 인물을 완전히 누락하였으며, 레퍼런스와 달리 물탱크가 옥탑 지붕 위에 잘못 배치되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "카메라 및 프레임 지시에 명시된 수리영의 흐릿한 어깨를 좌측 하단에 배치하여 오버 더 숄더 구도를 정확히 구현했으며, 구조물 레퍼런스의 디테일(조명 없음, 문 방향)을 잘 따랐습니다." }, { "label": "A", "score": 4, "verdict_ko": "구도 권한인 '카메라 및 프레임'에서 명시적으로 요구한 수리영의 어깨를 완전히 누락하여 지정된 레이아웃을 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "카메라 및 프레임 지시에 명시된 수리영의 흐릿한 어깨를 좌측 하단에 배치하여 오버 더 숄더 구도를 정확히 구현했으며, 구조물 레퍼런스의 디테일(조명 없음, 문 방향)을 잘 따랐습니다." }, { "label": "B", "score": 4, "verdict_ko": "구도 권한인 '카메라 및 프레임'에서 명시적으로 요구한 수리영의 어깨를 완전히 누락하여 지정된 레이아웃을 위반했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 8 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트+seed만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only", "plate_select": { "candidates": { "A": "L11B01", "B": "L11B02", "C": "L11B03", "D": "L11B04" }, "assigned": "L11B04", "choice": "Candidate D", "confident": true, "reason_ko": "\"옥탑방 현관 철문의 모습\"이라는 지문에 맞춰 옥탑방의 반쯤 열린 철문이 정면으로 보이는 서브공간입니다.", "kept": "L11B04" } }, "S12sh8::bgfirst_bg": { "input_fingerprint": "9418f463cfe309b0", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 젖혀진 커튼 뒤, 목을 앞으로 꺾은 채 바닥에 주저앉아 있는 강민숙의 피투성이 상반신.\n\nLOCATION (lock): Behind the curtain in the young woman's bedroom inside the rooftop home, dimly lit at night by the home's faint lamp light.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height just beyond the pulled curtain line, hold a slightly oblique medium frame on 강민숙's bloodied upper body, placing her collapsed form off-center while the curtain edge remains in the near foreground. Her dropped head and seated posture form a broken vertical, with the torn shoulder and collarbone region visible as the narrative center before the camera later tilts toward her hand.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 젖혀진 커튼 (Pulled aside); used as Creates a foreground reveal edge and partially encloses the body; 방 바닥의 혈흔 (Blood pooled on the floor); used as Grounds the body's seated position and carries evidence of the attack beneath it; 피 묻은 발자국 (Leading toward the curtain); used as Extends the discovered trail toward the curtain and the body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key room ambience reveals the bloodied body with muted neutrals and deep red accents, without assigning an unsupported bedroom light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 젖혀진 커튼 뒤, 목을 앞으로 꺾은 채 바닥에 주저앉아 있는 강민숙의 피투성이 상반신.\n\nLOCATION (lock): Behind the curtain in the young woman's bedroom inside the rooftop home, dimly lit at night by the home's faint lamp light.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height just beyond the pulled curtain line, hold a slightly oblique medium frame on 강민숙's bloodied upper body, placing her collapsed form off-center while the curtain edge remains in the near foreground. Her dropped head and seated posture form a broken vertical, with the torn shoulder and collarbone region visible as the narrative center before the camera later tilts toward her hand.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 젖혀진 커튼 (Pulled aside); used as Creates a foreground reveal edge and partially encloses the body; 방 바닥의 혈흔 (Blood pooled on the floor); used as Grounds the body's seated position and carries evidence of the attack beneath it; 피 묻은 발자국 (Leading toward the curtain); used as Extends the discovered trail toward the curtain and the body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key room ambience reveals the bloodied body with muted neutrals and deep red accents, without assigning an unsupported bedroom light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S12sh8__bgfirst_bg.png", "asset_id": "3f43168f-5656-4aac-80fb-aa25c5a6eec2", "input_asset_ids": [ "77040deb-09d7-499e-b45d-812ee2996e0d", "79e4504a-d22e-4b73-984b-2d62f766b914" ] }, "S12sh8": { "input_fingerprint": "ae7824ca32d309f4", "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): 젖혀진 커튼 뒤, 목을 앞으로 꺾은 채 바닥에 주저앉아 있는 강민숙의 피투성이 상반신.\n\nLOCATION (lock): Behind the curtain in the young woman's bedroom inside the rooftop home, dimly lit at night by the home's faint lamp light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height just beyond the pulled curtain line, hold a slightly oblique medium frame on 강민숙's bloodied upper body, placing her collapsed form off-center while the curtain edge remains in the near foreground. Her dropped head and seated posture form a broken vertical, with the torn shoulder and collarbone region visible as the narrative center before the camera later tilts toward her hand.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 젖혀진 커튼 (Pulled aside); used as Creates a foreground reveal edge and partially encloses the body; 방 바닥의 혈흔 (Blood pooled on the floor); used as Grounds the body's seated position and carries evidence of the attack beneath it; 피 묻은 발자국 (Leading toward the curtain); used as Extends the discovered trail toward the curtain and the body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key room ambience reveals the bloodied body with muted neutrals and deep red accents, without assigning an unsupported bedroom light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is seated motionless on the floor behind the curtain with her torso slumped and her head bent forward; one hand is closed around a crumpled photograph.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook remains dead, seated motionless on the floor behind the curtain with her torso slumped and her head bent forward amid the blood. Because she cannot grasp it later, one hand is already closed around a crumpled photograph, 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: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 젖혀진 커튼 뒤, 목을 앞으로 꺾은 채 바닥에 주저앉아 있는 강민숙의 피투성이 상반신.\n\nLOCATION (lock): Behind the curtain in the young woman's bedroom inside the rooftop home, dimly lit at night by the home's faint lamp light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height just beyond the pulled curtain line, hold a slightly oblique medium frame on 강민숙's bloodied upper body, placing her collapsed form off-center while the curtain edge remains in the near foreground. Her dropped head and seated posture form a broken vertical, with the torn shoulder and collarbone region visible as the narrative center before the camera later tilts toward her hand.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 젖혀진 커튼 (Pulled aside); used as Creates a foreground reveal edge and partially encloses the body; 방 바닥의 혈흔 (Blood pooled on the floor); used as Grounds the body's seated position and carries evidence of the attack beneath it; 피 묻은 발자국 (Leading toward the curtain); used as Extends the discovered trail toward the curtain and the body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key room ambience reveals the bloodied body with muted neutrals and deep red accents, without assigning an unsupported bedroom light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is seated motionless on the floor behind the curtain with her torso slumped and her head bent forward; one hand is closed around a crumpled photograph.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook remains dead, seated motionless on the floor behind the curtain with her torso slumped and her head bent forward amid the blood. Because she cannot grasp it later, one hand is already closed around a crumpled photograph, 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: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 젖혀진 커튼 뒤, 목을 앞으로 꺾은 채 바닥에 주저앉아 있는 강민숙의 피투성이 상반신.\n\nLOCATION (lock): Behind the curtain in the young woman's bedroom inside the rooftop home, dimly lit at night by the home's faint lamp light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From chest height just beyond the pulled curtain line, hold a slightly oblique medium frame on 강민숙's bloodied upper body, placing her collapsed form off-center while the curtain edge remains in the near foreground. Her dropped head and seated posture form a broken vertical, with the torn shoulder and collarbone region visible as the narrative center before the camera later tilts toward her hand.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 젖혀진 커튼 (Pulled aside); used as Creates a foreground reveal edge and partially encloses the body; 방 바닥의 혈흔 (Blood pooled on the floor); used as Grounds the body's seated position and carries evidence of the attack beneath it; 피 묻은 발자국 (Leading toward the curtain); used as Extends the discovered trail toward the curtain and the body.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key room ambience reveals the bloodied body with muted neutrals and deep red accents, without assigning an unsupported bedroom light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is seated motionless on the floor behind the curtain with her torso slumped and her head bent forward; one hand is closed around a crumpled photograph.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook remains dead, seated motionless on the floor behind the curtain with her torso slumped and her head bent forward amid the blood. Because she cannot grasp it later, one hand is already closed around a crumpled photograph, 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: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S12sh8__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S12sh8.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B02.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "지정된 미디엄 샷이 아닌 방 전체를 보여주는 와이드 샷이며, 캐릭터가 바닥이 아닌 침대에 앉아 있고 지정된 의상을 착용하지 않아 여러 지침을 위반했습니다." }, { "label": "B", "score": 9, "verdict_ko": "가슴 높이의 미디엄 샷 프레이밍, 전경의 커튼 배치, 바닥에 주저앉은 자세, 캐릭터의 레퍼런스 의상 및 사진 소품을 지시문대로 매우 충실하게 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "지정된 미디엄 샷이 아닌 방 전체를 보여주는 와이드 샷이며, 캐릭터가 바닥이 아닌 침대에 앉아 있고 지정된 의상을 착용하지 않아 여러 지침을 위반했습니다." }, { "label": "B", "score": 9, "verdict_ko": "가슴 높이의 미디엄 샷 프레이밍, 전경의 커튼 배치, 바닥에 주저앉은 자세, 캐릭터의 레퍼런스 의상 및 사진 소품을 지시문대로 매우 충실하게 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "바닥에 주저앉은 자세와 캐릭터의 외형을 잘 구현했으나, 지시된 상반신 미디엄 샷이 아닌 전신이 드러나는 넓은 프레이밍으로 연출된 점이 아쉽습니다." }, { "label": "B", "score": 2, "verdict_ko": "상반신 미디엄 샷 지시를 어겼을 뿐만 아니라, 바닥이 아닌 침대에 앉아 있고 하반신이 침구와 기형적으로 융합되는 치명적인 해부학적 오류(Hard Violation)가 있습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "바닥에 주저앉은 자세와 캐릭터의 외형을 잘 구현했으나, 지시된 상반신 미디엄 샷이 아닌 전신이 드러나는 넓은 프레이밍으로 연출된 점이 아쉽습니다." }, { "label": "A", "score": 2, "verdict_ko": "상반신 미디엄 샷 지시를 어겼을 뿐만 아니라, 바닥이 아닌 침대에 앉아 있고 하반신이 침구와 기형적으로 융합되는 치명적인 해부학적 오류(Hard Violation)가 있습니다." } ] }, "combined": { "totals": { "A": 4, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 4, "B": 16 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "지정된 미디엄 샷이 아닌 방 전체를 보여주는 와이드 샷이며, 캐릭터가 바닥이 아닌 침대에 앉아 있고 지정된 의상을 착용하지 않아 여러 지침을 위반했습니다." }, { "label": "B", "score": 9, "verdict_ko": "가슴 높이의 미디엄 샷 프레이밍, 전경의 커튼 배치, 바닥에 주저앉은 자세, 캐릭터의 레퍼런스 의상 및 사진 소품을 지시문대로 매우 충실하게 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B02.png" }, { "label": "CHARACTER REFERENCE — 강민숙: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "시신임에도 두 손으로 사진을 들고 있는 등 근육에 긴장이 들어간 능동적인 자세를 취하고 있습니다.", "fix_en": "Drop both hands heavily onto her lap so they show no muscular effort, leaving the photograph crumpled inside only one closed fist.", "unfixable": false }, { "issue_ko": "프롬프트에 명시된 손목의 작은 원형 자국(circular mark)이 보이지 않습니다.", "fix_en": "Add the required small circular mark onto her visible wrist.", "unfixable": false }, { "issue_ko": "상반신 중심의 미디엄 샷(medium shot)이어야 하나 하반신과 발까지 모두 프레임에 포함되어 있습니다.", "fix_en": "Reframe the camera to a medium shot focused tightly on her bloodied upper body.", "unfixable": true } ] }, "composition_critique": { "issues": [ { "issue_ko": "화면이 지나치게 넓어 강민숙의 다리 전체와 방의 넓은 영역까지 포함되며, 피투성이 상반신을 중심으로 한 미디엄 숏보다 인물이 작게 보인다.", "fix_en": "Crop tighter around Min-sook’s head, torso, and lap while retaining the near-foreground curtain edge and a small section of the bloodied floor.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Drop both hands heavily onto her lap so they show no muscular effort, leaving the photograph crumpled inside only one closed fist.\n- Add the required small circular mark onto her visible wrist.\n- Crop tighter around Min-sook’s head, torso, and lap while retaining the near-foreground curtain edge and a small section of the bloodied floor.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 손목의 작은 둥근 자국을 포함하였고 프레이밍이 요구사항에 조금 더 가깝습니다." }, { "label": "A", "score": 5, "verdict_ko": "손목에 명시된 둥근 자국이 누락되었으며 카메라 프레이밍이 지시된 미디엄 샷보다 더 넓게 잡혔습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 손목의 작은 둥근 자국을 포함하였고 프레이밍이 요구사항에 조금 더 가깝습니다." }, { "label": "A", "score": 5, "verdict_ko": "손목에 명시된 둥근 자국이 누락되었으며 카메라 프레이밍이 지시된 미디엄 샷보다 더 넓게 잡혔습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "중력에 따라 축 늘어진 시신의 자세를 사실적으로 구현했으며, 배경의 벽면 디테일(붉은 원 안의 사진)도 충실히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "의식이 남아 사진을 쳐다보는 듯한 목과 팔의 긴장이 느껴져 '무생물적 자세' 지침에 어긋나며, 벽면의 사진이 누락되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "중력에 따라 축 늘어진 시신의 자세를 사실적으로 구현했으며, 배경의 벽면 디테일(붉은 원 안의 사진)도 충실히 반영했습니다." }, { "label": "B", "score": 5, "verdict_ko": "의식이 남아 사진을 쳐다보는 듯한 목과 팔의 긴장이 느껴져 '무생물적 자세' 지침에 어긋나며, 벽면의 사진이 누락되었습니다." } ] }, "combined": { "totals": { "A": 12, "B": 12 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S12sh8__bgfirst_bg.png", "bg_asset_id": "3f43168f-5656-4aac-80fb-aa25c5a6eec2", "bg_record_key": "S12sh8::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B06", "choice": "Candidate F", "confident": true, "reason_ko": "지문에서 '젖혀진 커튼 뒤'라고만 명시되어 있어 특정 방을 구분하기 어려우므로, 커튼이 있는 방을 보여주는 현재 배정된 이미지를 유지합니다.", "kept": "L04B06" } }, "S12sh13::variants": { "author_fp": "666c47fc5963c718", "author": { "variants": [ { "approach_ko": "바닥과 거의 평행한 압축된 시선과 극도로 얕은 초점으로 굳은 손가락 사이의 사진 조각을 증거물처럼 고립시킨다.", "prompt_en": "Use a floor-skimming, tightly compressed close-up, with the fallen hand fully supported by the floor and the diagonal forearm receding softly toward the frame edge. Place critical focus on the locked, bloodied fingers and the small visible crease of the photographic face between them, letting the remaining hand and narrow floor plane fall into a very shallow, tactile blur. Restrained ambient light grazes the knuckles, dried and glossy blood, and crushed paper ridges, isolating the evidence with a controlled deep-red accent against muted darkness." }, { "approach_ko": "낮은 사선 원근과 더 깊은 초점으로 팔에서 손아귀와 사진까지 이어지는 물리적 연결과 피 묻은 바닥을 선명하게 드러낸다.", "prompt_en": "Photograph the close-up from a low oblique angle with a more immediate perspective and broader depth clarity, using the forearm’s diagonal as a strong visual path into the centered clenched hand and partly hidden photograph. Keep the hand visibly fallen and weight-bearing on the blood-marked floor, with the narrow surrounding plane rendered sharply enough to establish the crime-scene texture without competing with the evidence. Shape the dim ambient illumination across the floor and hand in restrained tonal layers, allowing the blood to carry the deepest red while the photograph’s creases catch a faint, crisp edge of light." } ] }, "reused": false }, "S12sh13": { "input_fingerprint": "b9f6397d5905b01d", "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): 피 묻은 강민숙의 굳은 손아귀 안에 구겨진 사진 한 장이 쥐어져 있는 클로즈업.\n\nLOCATION (lock): On the floor behind the bedroom curtain inside the rooftop home, beside the victim's body in the dim nighttime room. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the floor-level dolly-in beside 강민숙 with a close frame on her rigid, bloodied hand, allowing the hand and part of the forearm to provide human scale around the crumpled photograph. The photograph sits within the clenched fingers near the center but remains under forty percent of the image, while a narrow area of blood-marked floor preserves the crime-scene context.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 구겨진 사진 (Crumpled and held tightly) — A creased portion of the photographic face is visible between the fingers, but the depicted content is largely obscured by the clenched grip; used as The central evidentiary object trapped inside the dead woman's grip; 피 묻은 바닥 (Marked with blood); used as Provides a narrow contextual plane beneath the hand and photograph; 강민숙의 팔 (Rigid and bloodied) — The forearm enters diagonally from the frame edge toward the clenched hand; used as Connects the close detail to 강민숙's body outside the frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained low-key ambient illumination isolates the bloodied hand and crumpled photograph with a deep red accent against muted surroundings.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark bedroom lighting, curtain-side setting, blood state, and the dead woman's clothing and fixed seated position from the reference. Exclude the wider view of her face and torso, showing only the rigid hand and the crumpled photograph.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is seated motionless on the floor behind the curtain with her torso slumped and her head bent forward; one hand is closed around a crumpled photograph.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The crumpled photograph remains clenched in dead Min-sook’s bloodied, rigid hand until Su-ri-young removes it; she is seated motionless on the floor behind the curtain with her torso slumped and her head bent forward; one hand is closed around a crumpled photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 피 묻은 강민숙의 굳은 손아귀 안에 구겨진 사진 한 장이 쥐어져 있는 클로즈업.\n\nLOCATION (lock): On the floor behind the bedroom curtain inside the rooftop home, beside the victim's body in the dim nighttime room. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the floor-level dolly-in beside 강민숙 with a close frame on her rigid, bloodied hand, allowing the hand and part of the forearm to provide human scale around the crumpled photograph. The photograph sits within the clenched fingers near the center but remains under forty percent of the image, while a narrow area of blood-marked floor preserves the crime-scene context.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 구겨진 사진 (Crumpled and held tightly) — A creased portion of the photographic face is visible between the fingers, but the depicted content is largely obscured by the clenched grip; used as The central evidentiary object trapped inside the dead woman's grip; 피 묻은 바닥 (Marked with blood); used as Provides a narrow contextual plane beneath the hand and photograph; 강민숙의 팔 (Rigid and bloodied) — The forearm enters diagonally from the frame edge toward the clenched hand; used as Connects the close detail to 강민숙's body outside the frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained low-key ambient illumination isolates the bloodied hand and crumpled photograph with a deep red accent against muted surroundings.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark bedroom lighting, curtain-side setting, blood state, and the dead woman's clothing and fixed seated position from the reference. Exclude the wider view of her face and torso, showing only the rigid hand and the crumpled photograph.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is seated motionless on the floor behind the curtain with her torso slumped and her head bent forward; one hand is closed around a crumpled photograph.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The crumpled photograph remains clenched in dead Min-sook’s bloodied, rigid hand until Su-ri-young removes it; she is seated motionless on the floor behind the curtain with her torso slumped and her head bent forward; one hand is closed around a crumpled photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 floor-skimming, tightly compressed close-up, with the fallen hand fully supported by the floor and the diagonal forearm receding softly toward the frame edge. Place critical focus on the locked, bloodied fingers and the small visible crease of the photographic face between them, letting the remaining hand and narrow floor plane fall into a very shallow, tactile blur. Restrained ambient light grazes the knuckles, dried and glossy blood, and crushed paper ridges, isolating the evidence with a controlled deep-red accent against muted 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.", "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): 피 묻은 강민숙의 굳은 손아귀 안에 구겨진 사진 한 장이 쥐어져 있는 클로즈업.\n\nLOCATION (lock): On the floor behind the bedroom curtain inside the rooftop home, beside the victim's body in the dim nighttime room. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the floor-level dolly-in beside 강민숙 with a close frame on her rigid, bloodied hand, allowing the hand and part of the forearm to provide human scale around the crumpled photograph. The photograph sits within the clenched fingers near the center but remains under forty percent of the image, while a narrow area of blood-marked floor preserves the crime-scene context.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 구겨진 사진 (Crumpled and held tightly) — A creased portion of the photographic face is visible between the fingers, but the depicted content is largely obscured by the clenched grip; used as The central evidentiary object trapped inside the dead woman's grip; 피 묻은 바닥 (Marked with blood); used as Provides a narrow contextual plane beneath the hand and photograph; 강민숙의 팔 (Rigid and bloodied) — The forearm enters diagonally from the frame edge toward the clenched hand; used as Connects the close detail to 강민숙's body outside the frame.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Restrained low-key ambient illumination isolates the bloodied hand and crumpled photograph with a deep red accent against muted surroundings.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dark bedroom lighting, curtain-side setting, blood state, and the dead woman's clothing and fixed seated position from the reference. Exclude the wider view of her face and torso, showing only the rigid hand and the crumpled photograph.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): She is seated motionless on the floor behind the curtain with her torso slumped and her head bent forward; one hand is closed around a crumpled photograph.\n\nIMMOBILE BODIES OBEY GRAVITY: a person who is dead or unconscious\nexerts NO muscular effort. Every part of their body — head, torso,\narms, hands, fingers, legs — rests fully on whatever supports it\n(floor, wall, furniture, their own lap) and hangs or slumps with\ngravity. NEVER show any part of an immobile person's body lifted,\nraised, held up in the air, or posed as if presenting something:\nan object in their grip stays clenched in a hand that itself lies\nfallen on a support — the hand does not hold the object up. If the\ncanonical pose leaves a body part unspecified, resolve it as the\nmost gravity-compliant, fully-supported position.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The crumpled photograph remains clenched in dead Min-sook’s bloodied, rigid hand until Su-ri-young removes it; she is seated motionless on the floor behind the curtain with her torso slumped and her head bent forward; one hand is closed around a crumpled photograph.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 강민숙 (한국인 40대 초반 여성, 자연스러운 중년 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 close-up from a low oblique angle with a more immediate perspective and broader depth clarity, using the forearm’s diagonal as a strong visual path into the centered clenched hand and partly hidden photograph. Keep the hand visibly fallen and weight-bearing on the blood-marked floor, with the narrow surrounding plane rendered sharply enough to establish the crime-scene texture without competing with the evidence. Shape the dim ambient illumination across the floor and hand in restrained tonal layers, allowing the blood to carry the deepest red while the photograph’s creases catch a faint, crisp edge of light.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 7, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 클로즈업 프레이밍을 정확히 구현하여 상반신을 배제했으며, 바닥에 놓인 피 묻은 손과 구겨진 사진의 형태, 중력의 영향을 잘 표현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "얼굴과 상반신을 배제하라는 지시를 어겼으며, 앉아있어야 할 인물이 바닥에 누워 있어 고정된 자세 설정에 위배됩니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S12sh8_sel.png" }, { "label": "PROP REFERENCE — 피 묻은 낡은 어린아이 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "죽은 사람의 손가락과 손바닥이 중력을 거슬러 위로 구부러진 채 사진을 허공에 들고 있어, 근육에 힘이 들어간 것처럼 보입니다.", "fix_en": "Flatten the palm and fingers so they rest completely limply against the floor under the weight of gravity, with the photograph lying directly on the ground within the fallen grip.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Flatten the palm and fingers so they rest completely limply against the floor under the weight of gravity, with the photograph lying directly on the ground within the fallen grip.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "손아귀에 구겨진 사진을 꽉 쥐고 있는 모습을 프롬프트의 지시와 일치하게 잘 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "손을 펼친 채 사진 위에 얹어두어, 구겨진 사진을 꽉 쥐고 있어야 한다는 핵심 설정 및 포즈를 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "손아귀에 구겨진 사진을 꽉 쥐고 있는 모습을 프롬프트의 지시와 일치하게 잘 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "손을 펼친 채 사진 위에 얹어두어, 구겨진 사진을 꽉 쥐고 있어야 한다는 핵심 설정 및 포즈를 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 대로 피 묻은 손아귀에 구겨진 사진이 단단히 쥐어져 있으며, 바닥에 닿아 있는 무게감과 레퍼런스 속 소녀의 모습(스웨터)도 정확히 구현되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "손이 사진을 움켜쥐지 않고 펴진 채로 얹혀 있어 텍스트 묘사에 어긋나며, 사진 속 인물도 레퍼런스와 일치하지 않습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 대로 피 묻은 손아귀에 구겨진 사진이 단단히 쥐어져 있으며, 바닥에 닿아 있는 무게감과 레퍼런스 속 소녀의 모습(스웨터)도 정확히 구현되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "손이 사진을 움켜쥐지 않고 펴진 채로 얹혀 있어 텍스트 묘사에 어긋나며, 사진 속 인물도 레퍼런스와 일치하지 않습니다." } ] }, "combined": { "totals": { "A": 14, "B": 6 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S12sh8" }, "locked_char_refs_excluded": [ "강민숙(C04)" ] }, "S12sh17::variants": { "author_fp": "b731f8689a7b40bf", "author": { "variants": [ { "approach_ko": "정면성과 깊은 초점으로 빈 벽과 붉은 흔적을 냉정한 증거처럼 기록한다.", "prompt_en": "Use a restrained, observational lens feel with minimal perspective distortion and near-total depth of field. Keep the geometry quiet and frontal, rendering the wall texture, physical stroke edges, and lower bed edge with sober clarity; let the dim illumination fall evenly enough that the red mark reads as an unsettling piece of evidence within the surrounding negative space." }, { "approach_ko": "침대 가장자리를 부드럽게 흐리고 벽면의 질감과 끊긴 붉은 선에 불안을 집중시킨다.", "prompt_en": "Use a subtly compressed lens feel and selective focus, allowing the bed edge to fall gently soft while the wall surface and interrupted stroke remain sharply tactile. Shape the faint interior light with a slight raking quality across the wall so tiny surface irregularities and the stroke’s physical contact with the paint become ominously present, while the empty surrounding wall recedes into muted shadow." } ] }, "reused": false }, "S12sh17": { "input_fingerprint": "0782d14a023d2238", "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): 침대 위 빈 벽면, 붉은색 선이 반쯤 끊긴 미완성의 원형을 이루며 고정되어 있는 모습.\n\nLOCATION (lock): The bare wall directly above the bed in the young woman's bedroom inside the rooftop home, seen under faint nighttime interior light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold static at eye level near the side of the bed, facing the wall above it almost straight on in a medium environmental frame. Place the incomplete red circle in the upper-middle area and retain the bed edge along the lower frame as scale, leaving empty wall around the broken line so its interrupted shape reads immediately.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: incomplete red circle on wall in the upper-center of the frame, background; bed edge in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 침대 위 벽면 (Marked by a half-broken red circle) — The wall's room-facing side is shown nearly straight on, bearing a fixed red line that forms an incomplete circle; used as The supernatural focal surface carrying the interrupted circular mark; 침대 (Positioned below the marked wall) — Only its upper edge is visible from the side along the lower frame; used as Provides lower-frame spatial reference for the mark's position above it.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key ambient illumination holds the wall in muted tones while the explicitly red incomplete circle provides the sole strong color accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The photograph has been removed from Min-sook’s hand, while her wrist mark has turned red. Her seated corpse, the blood, and the newly forming red circle on the wall remain in the room.\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.\n\nSHOT TEXT (authoritative, Korean): 침대 위 빈 벽면, 붉은색 선이 반쯤 끊긴 미완성의 원형을 이루며 고정되어 있는 모습.\n\nLOCATION (lock): The bare wall directly above the bed in the young woman's bedroom inside the rooftop home, seen under faint nighttime interior light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold static at eye level near the side of the bed, facing the wall above it almost straight on in a medium environmental frame. Place the incomplete red circle in the upper-middle area and retain the bed edge along the lower frame as scale, leaving empty wall around the broken line so its interrupted shape reads immediately.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: incomplete red circle on wall in the upper-center of the frame, background; bed edge in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 침대 위 벽면 (Marked by a half-broken red circle) — The wall's room-facing side is shown nearly straight on, bearing a fixed red line that forms an incomplete circle; used as The supernatural focal surface carrying the interrupted circular mark; 침대 (Positioned below the marked wall) — Only its upper edge is visible from the side along the lower frame; used as Provides lower-frame spatial reference for the mark's position above it.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key ambient illumination holds the wall in muted tones while the explicitly red incomplete circle provides the sole strong color accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The photograph has been removed from Min-sook’s hand, while her wrist mark has turned red. Her seated corpse, the blood, and the newly forming red circle on the wall remain in the room.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, observational lens feel with minimal perspective distortion and near-total depth of field. Keep the geometry quiet and frontal, rendering the wall texture, physical stroke edges, and lower bed edge with sober clarity; let the dim illumination fall evenly enough that the red mark reads as an unsettling piece of evidence within the surrounding 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): night.\n\nSHOT TEXT (authoritative, Korean): 침대 위 빈 벽면, 붉은색 선이 반쯤 끊긴 미완성의 원형을 이루며 고정되어 있는 모습.\n\nLOCATION (lock): The bare wall directly above the bed in the young woman's bedroom inside the rooftop home, seen under faint nighttime interior light. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Hold static at eye level near the side of the bed, facing the wall above it almost straight on in a medium environmental frame. Place the incomplete red circle in the upper-middle area and retain the bed edge along the lower frame as scale, leaving empty wall around the broken line so its interrupted shape reads immediately.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: incomplete red circle on wall in the upper-center of the frame, background; bed edge in the lower-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 침대 위 벽면 (Marked by a half-broken red circle) — The wall's room-facing side is shown nearly straight on, bearing a fixed red line that forms an incomplete circle; used as The supernatural focal surface carrying the interrupted circular mark; 침대 (Positioned below the marked wall) — Only its upper edge is visible from the side along the lower frame; used as Provides lower-frame spatial reference for the mark's position above it.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key ambient illumination holds the wall in muted tones while the explicitly red incomplete circle provides the sole strong color accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The photograph has been removed from Min-sook’s hand, while her wrist mark has turned red. Her seated corpse, the blood, and the newly forming red circle on the wall remain in the room.\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 subtly compressed lens feel and selective focus, allowing the bed edge to fall gently soft while the wall surface and interrupted stroke remain sharply tactile. Shape the faint interior light with a slight raking quality across the wall so tiny surface irregularities and the stroke’s physical contact with the paint become ominously present, while the empty surrounding wall recedes into muted 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": { "B": 9, "A": 7 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "지시된 구도를 완벽히 따르며, 벽에 그려진 붉은 원이 그래픽 스티커가 아닌 실제 질감을 가진 흔적(혈흔)처럼 매우 사실적으로 묘사되었습니다." }, { "label": "A", "score": 7, "verdict_ko": "구도와 프레이밍은 지시사항을 잘 따랐으나, 벽에 그려진 붉은 원이 다소 인위적인 그래픽이나 테이프처럼 보여 질감의 사실감이 떨어집니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S12sh13_sel.png" } ], "critique": { "issues": [ { "issue_ko": "벽에 그려진 붉은색 원이 거의 다 이어져 있습니다. 제시문은 '반쯤 끊긴' 형태를 요구했습니다.", "fix_en": "Erase approximately half of the red circle's circumference so it forms a distinctly half-broken, incomplete ring.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Reduce the bed’s visible height to a narrow upper edge along the bottom of the frame, replacing the excess mattress and frame area with a continuation of the wall while preserving the circle and all other geometry.", "issue_ko": "침대가 하단 프레임의 좁은 윗선만 보여야 하는데 매트리스 측면과 침대 프레임이 크게 드러나 화면 하부를 과도하게 차지한다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Erase approximately half of the red circle's circumference so it forms a distinctly half-broken, incomplete ring.\n- Reduce the bed’s visible height to a narrow upper edge along the bottom of the frame, replacing the excess mattress and frame area with a continuation of the wall while preserving the circle and all other geometry.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 구도와 조명을 완벽히 따랐으며, 특히 벽면에 그려진 붉은색 선이 '반쯤 끊긴 미완성의 원형'이라는 핵심 지시를 매우 정확하게 시각화했습니다." }, { "label": "A", "score": 5, "verdict_ko": "전반적인 구도와 배경 설정은 지시를 잘 따랐으나, 붉은색 원의 끊어진 부분이 너무 작아 '반쯤 끊긴'이라는 텍스트 묘사를 충실히 반영하지 못했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 구도와 조명을 완벽히 따랐으며, 특히 벽면에 그려진 붉은색 선이 '반쯤 끊긴 미완성의 원형'이라는 핵심 지시를 매우 정확하게 시각화했습니다." }, { "label": "A", "score": 5, "verdict_ko": "전반적인 구도와 배경 설정은 지시를 잘 따랐으나, 붉은색 원의 끊어진 부분이 너무 작아 '반쯤 끊긴'이라는 텍스트 묘사를 충실히 반영하지 못했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 '반쯤 끊긴 미완성의 원형'을 벽면에 매우 정확하게 구현하였으며, 요구된 구도와 조명 분위기를 훌륭하게 따랐습니다." }, { "label": "B", "score": 5, "verdict_ko": "전반적인 구도와 분위기는 맞지만, 붉은색 원이 거의 연결되어 있어 '반쯤 끊긴'이라는 핵심적인 지시사항을 제대로 반영하지 못했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 '반쯤 끊긴 미완성의 원형'을 벽면에 매우 정확하게 구현하였으며, 요구된 구도와 조명 분위기를 훌륭하게 따랐습니다." }, { "label": "A", "score": 5, "verdict_ko": "전반적인 구도와 분위기는 맞지만, 붉은색 원이 거의 연결되어 있어 '반쯤 끊긴'이라는 핵심적인 지시사항을 제대로 반영하지 못했습니다." } ] }, "combined": { "totals": { "A": 10, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev만 (배경 전용·공유 계획)", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S12sh13" }, "lane_policy": "share_plan_prev_bgonly", "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B06", "choice": "Candidate F", "confident": true, "reason_ko": "\"침대 위 빈 벽면, 붉은색 선이 반쯤 끊긴 미완성의 원형\"이라는 묘사와 침대 위 벽면에 그려진 붉은 원형 자국이 정확히 일치합니다.", "kept": "L04B06" } }, "S13sh1::variants": { "author_fp": "afb6e58f33b2d7e3", "author": { "variants": [ { "approach_ko": "얼굴은 선명하게 붙잡고 팔다리와 옷자락의 미세한 흐림으로 공포에 찬 도주 속도를 체감시키는 밀착 추적 촬영.", "prompt_en": "Photograph this as a visceral close-tracking medium shot, keeping Suriyeong’s tear-soaked face and straining upper torso crisply legible while allowing restrained directional motion blur along her clothing, extended leg, and suspended foot. Use shallow depth separation so the entrance recedes into darkness and the steep stair line remains readable without competing with her expression. Shape the low-key night light across the wet tear tracks with a narrow, soft glint, preserving natural skin texture and a breathless sense of forward momentum." }, { "approach_ko": "계단의 가파른 하강선과 현관의 어두운 틀을 또렷하게 살려 수리영이 위험한 공간 속으로 내던져지는 순간을 강조한 촬영.", "prompt_en": "Photograph this as a tense architectural medium shot with greater depth clarity, holding Suriyeong, the half-open entrance behind her, and the descending stair geometry in sharp spatial relation. Emphasize the strong diagonal beneath her airborne foot and use the doorway as a dark enclosing frame around the route she has just fled, making her dynamic posture feel exposed against the rigid structure. Keep motion nearly frozen and render the muted night contrast with firm shadow edges and controlled highlights on tears, metal, and worn surfaces." } ] }, "reused": false }, "S13sh1": { "input_fingerprint": "3b1f0ce0132b6814", "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 옥탑방 현관 밖, 눈물이 범벅된 얼굴로 가파른 계단을 향해 한쪽 발을 허공에 뻗은 채 뛰어내리는 역동적 자세의 수리영의 상반신.\n\nLOCATION (lock): On the steep exterior stairway immediately outside the rooftop home's entrance at night. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just outside the rooftop-room entrance at Suriyeong’s upper-torso height, the camera holds close beside the top of the steep stairs as she bursts through in a three-quarter view. Her torso occupies the center-left while her extended leg and suspended foot remain visible in the lower-right, clearly committing her movement down the stairs as the camera begins to track beside her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, moves toward descending staircase; descending staircase in the lower-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 입구 (In use as her escape route) — The exterior-facing side and passage are visible behind Suriyeong; used as The entrance frames the point from which Suriyeong has just escaped; 가파른 계단 (Unobstructed) — The descending run angles from the middle of the frame toward the lower-right; used as Provides the strong downward line beneath her suspended foot.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dark night ambience is rendered with low-key contrast, a restrained muted-neutral grade, and tactile detail retained on Suriyeong’s tear-covered face.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 steep rooftop stair access, metal entrance door, deep night lighting, and the door's half-open state from the reference. Exclude the earlier empty doorway composition and any courtyard, bicycle, or police vehicle not visible from this upper stair 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): As Su-ri-young flees, Min-sook’s seated corpse, the bloodstains, and the red wall circle remain inside the rooftop room until their later unexplained removal.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 옥탑방 현관 밖, 눈물이 범벅된 얼굴로 가파른 계단을 향해 한쪽 발을 허공에 뻗은 채 뛰어내리는 역동적 자세의 수리영의 상반신.\n\nLOCATION (lock): On the steep exterior stairway immediately outside the rooftop home's entrance at night. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just outside the rooftop-room entrance at Suriyeong’s upper-torso height, the camera holds close beside the top of the steep stairs as she bursts through in a three-quarter view. Her torso occupies the center-left while her extended leg and suspended foot remain visible in the lower-right, clearly committing her movement down the stairs as the camera begins to track beside her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, moves toward descending staircase; descending staircase in the lower-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 입구 (In use as her escape route) — The exterior-facing side and passage are visible behind Suriyeong; used as The entrance frames the point from which Suriyeong has just escaped; 가파른 계단 (Unobstructed) — The descending run angles from the middle of the frame toward the lower-right; used as Provides the strong downward line beneath her suspended foot.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dark night ambience is rendered with low-key contrast, a restrained muted-neutral grade, and tactile detail retained on Suriyeong’s tear-covered face.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 steep rooftop stair access, metal entrance door, deep night lighting, and the door's half-open state from the reference. Exclude the earlier empty doorway composition and any courtyard, bicycle, or police vehicle not visible from this upper stair 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): As Su-ri-young flees, Min-sook’s seated corpse, the bloodstains, and the red wall circle remain inside the rooftop room until their later unexplained removal.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 visceral close-tracking medium shot, keeping Suriyeong’s tear-soaked face and straining upper torso crisply legible while allowing restrained directional motion blur along her clothing, extended leg, and suspended foot. Use shallow depth separation so the entrance recedes into darkness and the steep stair line remains readable without competing with her expression. Shape the low-key night light across the wet tear tracks with a narrow, soft glint, preserving natural skin texture and a breathless sense of forward momentum.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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 옥탑방 현관 밖, 눈물이 범벅된 얼굴로 가파른 계단을 향해 한쪽 발을 허공에 뻗은 채 뛰어내리는 역동적 자세의 수리영의 상반신.\n\nLOCATION (lock): On the steep exterior stairway immediately outside the rooftop home's entrance at night. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just outside the rooftop-room entrance at Suriyeong’s upper-torso height, the camera holds close beside the top of the steep stairs as she bursts through in a three-quarter view. Her torso occupies the center-left while her extended leg and suspended foot remain visible in the lower-right, clearly committing her movement down the stairs as the camera begins to track beside her.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, moves toward descending staircase; descending staircase in the lower-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥탑방 입구 (In use as her escape route) — The exterior-facing side and passage are visible behind Suriyeong; used as The entrance frames the point from which Suriyeong has just escaped; 가파른 계단 (Unobstructed) — The descending run angles from the middle of the frame toward the lower-right; used as Provides the strong downward line beneath her suspended foot.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dark night ambience is rendered with low-key contrast, a restrained muted-neutral grade, and tactile detail retained on Suriyeong’s tear-covered face.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 steep rooftop stair access, metal entrance door, deep night lighting, and the door's half-open state from the reference. Exclude the earlier empty doorway composition and any courtyard, bicycle, or police vehicle not visible from this upper stair 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): As Su-ri-young flees, Min-sook’s seated corpse, the bloodstains, and the red wall circle remain inside the rooftop room until their later unexplained removal.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 tense architectural medium shot with greater depth clarity, holding Suriyeong, the half-open entrance behind her, and the descending stair geometry in sharp spatial relation. Emphasize the strong diagonal beneath her airborne foot and use the doorway as a dark enclosing frame around the route she has just fled, making her dynamic posture feel exposed against the rigid structure. Keep motion nearly frozen and render the muted night contrast with firm shadow edges and controlled highlights on tears, metal, and worn surfaces.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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 옥상 콘크리트 구조와 초록색 바닥을 잘 유지했으며, 지시된 미디엄 샷 프레이밍과 뛰어내리는 역동적인 포즈를 훌륭하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "레퍼런스에 존재하지 않는 철제 발코니와 외부 계단을 완전히 새롭게 창조해 내어 장소 일관성을 심각하게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S11sh3_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "프롬프트는 수리영의 상반신을 중심으로 한 미디엄 샷을 지시했으나, 머리부터 발끝까지 전신이 모두 프레임에 들어오는 롱 샷(풀 샷)으로 렌더링되었습니다.", "fix_en": "Change the camera scale to a medium shot focusing on Suriyeong's upper torso.", "unfixable": true }, { "issue_ko": "의상, 뻗은 다리, 공중에 뜬 발에 요구된 방향성 모션 블러(directional motion blur) 효과가 생략되어 움직임이 선명하게만 묘사되었습니다.", "fix_en": "Apply directional motion blur along her clothing, extended right leg, and suspended foot to convey rapid forward movement.", "unfixable": false }, { "issue_ko": "캐릭터의 오른손 손가락 형태가 뭉개져 있으며, 문 가장자리와 융합되어 해부학적으로 부자연스럽습니다.", "fix_en": "Reconstruct the right hand to feature clearly defined, anatomically correct fingers cleanly separated from the door's edge.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Shift Suriyeong slightly left within the frame while preserving her pose, scale, and the suspended foot in the lower-right.", "issue_ko": "수리영의 상반신이 지정된 화면 중간 왼쪽이 아니라 거의 정중앙에 배치되어 있다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Apply directional motion blur along her clothing, extended right leg, and suspended foot to convey rapid forward movement.\n- Reconstruct the right hand to feature clearly defined, anatomically correct fingers cleanly separated from the door's edge.\n- Shift Suriyeong slightly left within the frame while preserving her pose, scale, and the suspended foot in the lower-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.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "프롬프트가 요구한 의상과 뻗은 다리의 방향성 모션 블러(motion blur)를 정확히 적용하여 역동적인 하강 순간을 훌륭하게 연출했습니다." }, { "label": "A", "score": 7, "verdict_ko": "구도와 배경, 인물의 정체성은 잘 맞췄으나, 지시사항에 명시된 모션 블러 효과가 완전히 누락되어 정적인 이미지가 되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "프롬프트가 요구한 의상과 뻗은 다리의 방향성 모션 블러(motion blur)를 정확히 적용하여 역동적인 하강 순간을 훌륭하게 연출했습니다." }, { "label": "A", "score": 7, "verdict_ko": "구도와 배경, 인물의 정체성은 잘 맞췄으나, 지시사항에 명시된 모션 블러 효과가 완전히 누락되어 정적인 이미지가 되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "얼굴과 상반신까지 과도한 모션 블러가 적용되어 눈물 젖은 얼굴을 선명하게 유지하라는 지시를 위반했습니다." }, { "label": "B", "score": 10, "verdict_ko": "얼굴과 상반신은 선명하게 유지하면서 다리와 발에만 절제된 모션 블러를 적용하여 프롬프트의 시각적 접근 방식을 완벽하게 구현했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "얼굴과 상반신까지 과도한 모션 블러가 적용되어 눈물 젖은 얼굴을 선명하게 유지하라는 지시를 위반했습니다." }, { "label": "A", "score": 10, "verdict_ko": "얼굴과 상반신은 선명하게 유지하면서 다리와 발에만 절제된 모션 블러를 적용하여 프롬프트의 시각적 접근 방식을 완벽하게 구현했습니다." } ] }, "combined": { "totals": { "A": 17, "B": 15 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S11sh3" }, "lane_policy": "ab_select_bypass:prev" }, "S13sh5::bgfirst_bg": { "input_fingerprint": "267e036148ac6cc5", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 혜수의 가슴팍 부근을 올려다보며 뺨에 눈물이 고여 흐르고 있는 수리영의 얼굴.\n\nLOCATION (lock): In the outdoor yard or alley at the base of the multi-unit villa, near the arriving vehicle and rooftop stairs at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The dolly-in finishes close to Hyesu’s chest position and just above Suriyeong’s eye line, looking gently downward while preserving Suriyeong’s physically natural upward eyeline. Her tearful face occupies most of the center frame, turned slightly three-quarter rather than square to camera, with only a narrow trace of the exterior setting left at the edge.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, looks toward Hyesu above her eyeline.\n- KEY BACKGROUND ELEMENTS: 계단 방향 (Unobstructed) — Only the side of the stair approach remains near the frame edge; used as A narrow peripheral spatial cue maintains continuity with Suriyeong’s descent and stop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key night ambience and a muted-neutral grade hold detail in the tears without introducing a new visible light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 혜수의 가슴팍 부근을 올려다보며 뺨에 눈물이 고여 흐르고 있는 수리영의 얼굴.\n\nLOCATION (lock): In the outdoor yard or alley at the base of the multi-unit villa, near the arriving vehicle and rooftop stairs at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The dolly-in finishes close to Hyesu’s chest position and just above Suriyeong’s eye line, looking gently downward while preserving Suriyeong’s physically natural upward eyeline. Her tearful face occupies most of the center frame, turned slightly three-quarter rather than square to camera, with only a narrow trace of the exterior setting left at the edge.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, looks toward Hyesu above her eyeline.\n- KEY BACKGROUND ELEMENTS: 계단 방향 (Unobstructed) — Only the side of the stair approach remains near the frame edge; used as A narrow peripheral spatial cue maintains continuity with Suriyeong’s descent and stop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key night ambience and a muted-neutral grade hold detail in the tears without introducing a new visible light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S13sh5__bgfirst_bg.png", "asset_id": "1b9f6c36-1434-4246-a511-6df597d11914", "input_asset_ids": [ "caffda7c-d779-4236-90b4-f4c85537c58b", "d8e77c34-2e1e-4ba7-8c80-b4d48d398fe6" ] }, "S13sh5": { "input_fingerprint": "57a765537c2eaa3c", "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): 혜수의 가슴팍 부근을 올려다보며 뺨에 눈물이 고여 흐르고 있는 수리영의 얼굴.\n\nLOCATION (lock): In the outdoor yard or alley at the base of the multi-unit villa, near the arriving vehicle and rooftop stairs at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The dolly-in finishes close to Hyesu’s chest position and just above Suriyeong’s eye line, looking gently downward while preserving Suriyeong’s physically natural upward eyeline. Her tearful face occupies most of the center frame, turned slightly three-quarter rather than square to camera, with only a narrow trace of the exterior setting left at the edge.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, looks toward Hyesu above her eyeline.\n- KEY BACKGROUND ELEMENTS: 계단 방향 (Unobstructed) — Only the side of the stair approach remains near the frame edge; used as A narrow peripheral spatial cue maintains continuity with Suriyeong’s descent and stop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key night ambience and a muted-neutral grade hold detail in the tears without introducing a new visible light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook’s corpse and the bloody room are still presumed to be upstairs at this point, before Hye-soo and Detective Kim enter and discover that everything has vanished.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수의 가슴팍 부근을 올려다보며 뺨에 눈물이 고여 흐르고 있는 수리영의 얼굴.\n\nLOCATION (lock): In the outdoor yard or alley at the base of the multi-unit villa, near the arriving vehicle and rooftop stairs at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The dolly-in finishes close to Hyesu’s chest position and just above Suriyeong’s eye line, looking gently downward while preserving Suriyeong’s physically natural upward eyeline. Her tearful face occupies most of the center frame, turned slightly three-quarter rather than square to camera, with only a narrow trace of the exterior setting left at the edge.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, looks toward Hyesu above her eyeline.\n- KEY BACKGROUND ELEMENTS: 계단 방향 (Unobstructed) — Only the side of the stair approach remains near the frame edge; used as A narrow peripheral spatial cue maintains continuity with Suriyeong’s descent and stop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key night ambience and a muted-neutral grade hold detail in the tears without introducing a new visible light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook’s corpse and the bloody room are still presumed to be upstairs at this point, before Hye-soo and Detective Kim enter and discover that everything has vanished.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 혜수의 가슴팍 부근을 올려다보며 뺨에 눈물이 고여 흐르고 있는 수리영의 얼굴.\n\nLOCATION (lock): In the outdoor yard or alley at the base of the multi-unit villa, near the arriving vehicle and rooftop stairs at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The dolly-in finishes close to Hyesu’s chest position and just above Suriyeong’s eye line, looking gently downward while preserving Suriyeong’s physically natural upward eyeline. Her tearful face occupies most of the center frame, turned slightly three-quarter rather than square to camera, with only a narrow trace of the exterior setting left at the edge.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, foreground, looks toward Hyesu above her eyeline.\n- KEY BACKGROUND ELEMENTS: 계단 방향 (Unobstructed) — Only the side of the stair approach remains near the frame edge; used as A narrow peripheral spatial cue maintains continuity with Suriyeong’s descent and stop.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key night ambience and a muted-neutral grade hold detail in the tears without introducing a new visible light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook’s corpse and the bloody room are still presumed to be upstairs at this point, before Hye-soo and Detective Kim enter and discover that everything has vanished.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S13sh5__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S13sh5.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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L11B02.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" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "수리영의 눈물 젖은 얼굴이 화면 대부분을 채우는 타이트한 클로즈업과 상대방 가슴 높이에서 약간 내려다보는 카메라 시점을 정확히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 클로즈업 비율을 벗어나 앵글이 넓어지면서 배경의 골목길과 주변 환경이 프레임에 과도하게 노출되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "수리영의 눈물 젖은 얼굴이 화면 대부분을 채우는 타이트한 클로즈업과 상대방 가슴 높이에서 약간 내려다보는 카메라 시점을 정확히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 클로즈업 비율을 벗어나 앵글이 넓어지면서 배경의 골목길과 주변 환경이 프레임에 과도하게 노출되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 클로즈업 프레이밍을 정확히 구현하여 인물의 얼굴이 화면 대부분을 차지하고 배경 노출이 최소화됨." }, { "label": "B", "score": 4, "verdict_ko": "요구된 타이트한 클로즈업 샷에 비해 프레임이 너무 넓어 주변 배경과 거리가 과도하게 드러남." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 클로즈업 프레이밍을 정확히 구현하여 인물의 얼굴이 화면 대부분을 차지하고 배경 노출이 최소화됨." }, { "label": "A", "score": 4, "verdict_ko": "요구된 타이트한 클로즈업 샷에 비해 프레임이 너무 넓어 주변 배경과 거리가 과도하게 드러남." } ] }, "combined": { "totals": { "A": 8, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 8, "B": 14 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "수리영의 눈물 젖은 얼굴이 화면 대부분을 채우는 타이트한 클로즈업과 상대방 가슴 높이에서 약간 내려다보는 카메라 시점을 정확히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 클로즈업 비율을 벗어나 앵글이 넓어지면서 배경의 골목길과 주변 환경이 프레임에 과도하게 노출되었습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L11B02.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [] }, "composition_critique": { "issues": [ { "fix_en": "Remove the large foreground torso from the right side and extend the existing peripheral exterior background into that area, keeping Suriyeong’s face dominant and centered.", "issue_ko": "오른쪽의 큰 전경 인물 몸통이 화면을 과도하게 차지해 수리영의 얼굴 중심 구도와 가장자리에 외부 공간의 좁은 흔적만 남겨야 한다는 구성을 방해한다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the large foreground torso from the right side and extend the existing peripheral exterior background into that area, keeping Suriyeong’s face dominant and centered.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 카메라 위치와 시선을 반영하여 전경에 혜수의 가슴팍을 배치함으로써 지문이 요구하는 공간감과 구도를 매우 정확히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "수리영의 표정과 눈물은 잘 표현되었으나, 카메라 위치와 시선의 기준이 되는 전경 요소(혜수)가 누락되어 샷의 디테일이 부족합니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 카메라 위치와 시선을 반영하여 전경에 혜수의 가슴팍을 배치함으로써 지문이 요구하는 공간감과 구도를 매우 정확히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "수리영의 표정과 눈물은 잘 표현되었으나, 카메라 위치와 시선의 기준이 되는 전경 요소(혜수)가 누락되어 샷의 디테일이 부족합니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "수리영을 전경 및 중앙에 배치하고 지시된 카메라 앵글과 시선 방향을 정확하게 구현함." }, { "label": "B", "score": 5, "verdict_ko": "수리영이 전경에 있어야 한다는 지시를 어기고 화면 우측 전경에 프레임을 크게 가리는 피사체를 배치함." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "수리영을 전경 및 중앙에 배치하고 지시된 카메라 앵글과 시선 방향을 정확하게 구현함." }, { "label": "A", "score": 5, "verdict_ko": "수리영이 전경에 있어야 한다는 지시를 어기고 화면 우측 전경에 프레임을 크게 가리는 피사체를 배치함." } ] }, "combined": { "totals": { "A": 12, "B": 13 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S13sh5__bgfirst_bg.png", "bg_asset_id": "1b9f6c36-1434-4246-a511-6df597d11914", "bg_record_key": "S13sh5::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L11B01", "B": "L11B02", "C": "L11B03", "D": "L11B04" }, "assigned": "L11B02", "choice": "Candidate B", "confident": false, "reason_ko": "지문이 인물의 감정과 얼굴 클로즈업만을 묘사하고 있어 건물 주변, 계단, 옥상 중 특정 하위 공간을 구분할 단서가 없으므로 기존 할당을 유지함.", "kept": "L11B02" } }, "S14sh4::variants": { "author_fp": "97f64314b8f44be5", "author": { "variants": [ { "approach_ko": "정중앙의 원근과 깊은 초점으로 방 전체를 냉정한 증거 사진처럼 제시한다.", "prompt_en": "Use a formal, centered one-point composition with a restrained broad-angle lens feel and clean rectilinear geometry. Hold the doorway, middle distance, and rear plane in deep focus, giving every surface equal evidentiary clarity. Keep the ambient night interior light soft and clinically even within restrained low-key contrast, emphasizing unnerving order through symmetry, stillness, and muted neutral texture." }, { "approach_ko": "어두운 문틀 전경과 압축된 공간 층위를 활용해 누군가 문턱에서 숨죽여 확인하는 듯한 불안을 만든다.", "prompt_en": "Use a gently compressed lens feel, letting the near doorway edges read as dark foreground occlusion around the illuminated room. Arrange the bed, floor, and curtain area as tightly stacked depth planes, with subtle focus falloff that preserves clear readability throughout. Shape the ambient night illumination with a delicate lateral tonal gradient and deeper corner shadows, creating a tense, watchful reveal rather than clinical symmetry." } ] }, "reused": false }, "S14sh4": { "input_fingerprint": "02178239a94c97ef", "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): 열린 방문 너머, 핏자국 하나 없이 깨끗하게 정돈된 침대와 바닥의 전경.\n\nLOCATION (lock): The young woman's bedroom inside the rooftop home, now unnaturally clean and orderly under nighttime interior light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera pauses at eye level in the open doorway, far enough back to present the bed and visible floor together in an unobstructed wide composition. The doorway edges contain the empty room while the clean bed, bare floor, and curtain area remain equally readable, making the complete absence of blood, the red mark, and the body the visual event.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 열린 방문 (Open) — The doorway opening faces the camera and exposes the room beyond; used as Creates a fixed architectural frame around the room reveal; 침대 (Clean and excessively tidy; no red mark visible) — The top and camera-facing side are both visible from the doorway; used as Primary evidence area held clearly in the middle distance; 바닥 (Clean; no bloodstain visible); used as Keeps the absence of any trace legible across the lower half of the room; 커튼과 그 뒤 공간 (No body present); used as Marks the area where the body had been expected.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior illumination appropriate to the night setting is treated with restrained low-key contrast and muted neutrals, without assigning an unshown source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bedroom has been reset to an excessively clean state: Min-sook’s corpse, all blood, and the red wall circle are gone.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 열린 방문 너머, 핏자국 하나 없이 깨끗하게 정돈된 침대와 바닥의 전경.\n\nLOCATION (lock): The young woman's bedroom inside the rooftop home, now unnaturally clean and orderly under nighttime interior light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera pauses at eye level in the open doorway, far enough back to present the bed and visible floor together in an unobstructed wide composition. The doorway edges contain the empty room while the clean bed, bare floor, and curtain area remain equally readable, making the complete absence of blood, the red mark, and the body the visual event.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 열린 방문 (Open) — The doorway opening faces the camera and exposes the room beyond; used as Creates a fixed architectural frame around the room reveal; 침대 (Clean and excessively tidy; no red mark visible) — The top and camera-facing side are both visible from the doorway; used as Primary evidence area held clearly in the middle distance; 바닥 (Clean; no bloodstain visible); used as Keeps the absence of any trace legible across the lower half of the room; 커튼과 그 뒤 공간 (No body present); used as Marks the area where the body had been expected.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior illumination appropriate to the night setting is treated with restrained low-key contrast and muted neutrals, without assigning an unshown source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bedroom has been reset to an excessively clean state: Min-sook’s corpse, all blood, and the red wall circle are gone.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a formal, centered one-point composition with a restrained broad-angle lens feel and clean rectilinear geometry. Hold the doorway, middle distance, and rear plane in deep focus, giving every surface equal evidentiary clarity. Keep the ambient night interior light soft and clinically even within restrained low-key contrast, emphasizing unnerving order through symmetry, stillness, and muted neutral texture.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 열린 방문 너머, 핏자국 하나 없이 깨끗하게 정돈된 침대와 바닥의 전경.\n\nLOCATION (lock): The young woman's bedroom inside the rooftop home, now unnaturally clean and orderly under nighttime interior light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The camera pauses at eye level in the open doorway, far enough back to present the bed and visible floor together in an unobstructed wide composition. The doorway edges contain the empty room while the clean bed, bare floor, and curtain area remain equally readable, making the complete absence of blood, the red mark, and the body the visual event.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 열린 방문 (Open) — The doorway opening faces the camera and exposes the room beyond; used as Creates a fixed architectural frame around the room reveal; 침대 (Clean and excessively tidy; no red mark visible) — The top and camera-facing side are both visible from the doorway; used as Primary evidence area held clearly in the middle distance; 바닥 (Clean; no bloodstain visible); used as Keeps the absence of any trace legible across the lower half of the room; 커튼과 그 뒤 공간 (No body present); used as Marks the area where the body had been expected.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior illumination appropriate to the night setting is treated with restrained low-key contrast and muted neutrals, without assigning an unshown source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bedroom has been reset to an excessively clean state: Min-sook’s corpse, all blood, and the red wall circle are gone.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a gently compressed lens feel, letting the near doorway edges read as dark foreground occlusion around the illuminated room. Arrange the bed, floor, and curtain area as tightly stacked depth planes, with subtle focus falloff that preserves clear readability throughout. Shape the ambient night illumination with a delicate lateral tonal gradient and deeper corner shadows, creating a tense, watchful reveal rather than clinical symmetry.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": "레퍼런스의 소품(토끼 인형, 가방)과 벽지 질감을 정확히 살렸으며, 열린 방문 너머의 정돈된 방이라는 프롬프트의 구도를 훌륭하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "레퍼런스에 없는 나무 패널 벽과 우측의 추가 통로를 임의로 생성하여 로케이션 설정(Location lock)을 크게 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B07.png" } ], "critique": { "issues": [ { "issue_ko": "전경 왼쪽의 문이 레퍼런스 이미지의 평평한 미닫이문 틀 구조와 다르게 손잡이가 달린 패널형 여닫이문으로 묘사되었습니다.", "fix_en": "Remove the doorknob, panels, and grooved details from the dark door on the left foreground, turning it into a flat, plain vertical wall edge matching the sliding doorway architecture in the reference image.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "issue_ko": "반쯤 열린 문짝이 화면 왼쪽의 큰 전경 면적을 차지해 출입구 너머 방을 가리지 않는 와이드 구도를 방해한다.", "fix_en": "Swing the door fully open out of the frame so the doorway edges frame an unobstructed wide view of the room, preserving the existing camera position and room geometry.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the doorknob, panels, and grooved details from the dark door on the left foreground, turning it into a flat, plain vertical wall edge matching the sliding doorway architecture in the reference image.\n- Swing the door fully open out of the frame so the doorway edges frame an unobstructed wide view of the room, preserving the existing camera position and room geometry.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 앵글과 방의 청결한 상태를 충실히 구현했으며, 전경에 배치된 문의 물리적 구조도 자연스럽습니다." }, { "label": "B", "score": 3, "verdict_ko": "왼쪽 전경에 문의 얇은 측면에 둥근 문고리가 달려 있는 물리적으로 불가능한 구조가 생성되어 최우선 규칙을 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 앵글과 방의 청결한 상태를 충실히 구현했으며, 전경에 배치된 문의 물리적 구조도 자연스럽습니다." }, { "label": "B", "score": 3, "verdict_ko": "왼쪽 전경에 문의 얇은 측면에 둥근 문고리가 달려 있는 물리적으로 불가능한 구조가 생성되어 최우선 규칙을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 구도에 맞춰 핏자국 없이 정돈된 방을 잘 표현했으며, '어두운 전경 가림막(dark foreground occlusion)' 지침을 충실히 따랐습니다." }, { "label": "B", "score": 6, "verdict_ko": "깨끗한 방의 전경을 잘 보여주나, 화면 왼쪽의 열린 문이 밝게 표현되어 전경을 어둡게 가리라는 시각적 접근 지침에서 다소 벗어납니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 구도에 맞춰 핏자국 없이 정돈된 방을 잘 표현했으며, '어두운 전경 가림막(dark foreground occlusion)' 지침을 충실히 따랐습니다." }, { "label": "A", "score": 6, "verdict_ko": "깨끗한 방의 전경을 잘 보여주나, 화면 왼쪽의 열린 문이 밝게 표현되어 전경을 어둡게 가리라는 시각적 접근 지침에서 다소 벗어납니다." } ] }, "combined": { "totals": { "A": 13, "B": 10 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B07", "choice": "Candidate G", "confident": true, "reason_ko": "\"열린 방문 너머\"와 \"깨끗하게 정돈된 침대와 바닥\"이라는 묘사에 정확히 부합하며, 야간 조명 설정도 일치함.", "kept": "L04B07" } }, "S14sh8::bgfirst_bg": { "input_fingerprint": "026961cdc6b6c4b8", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 식탁 위로 두 손을 뻗은 채 허공을 가리키고 있는 수리영의 절박한 표정.\n\nLOCATION (lock): The dining area of the rooftop home's living room, centered on the now-clean table under the home's nighttime lighting.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Tracking around the near side of the dining table settles at chest height into a close-medium, nearly frontal view justified by Suriyeong’s desperate attempt to prove what she saw. Her face holds the upper center while both extended hands enter the lower foreground above the clean tabletop, indicating two separate empty places before the camera continues laterally.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: empty dining tabletop beneath Suriyeong's hands in the lower-center of the frame, foreground; 수리영 in the middle-center of the frame, midground, points to two empty places on the table.\n- KEY BACKGROUND ELEMENTS: 식탁 (Clean; two claimed teacups absent) — The top plane angles toward the camera beneath her hands; used as Supports Suriyeong’s extended hands and exposes the two places where she says cups had been.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral interior ambience is shaped into moody low-key contrast with a restrained, tactile muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 식탁 위로 두 손을 뻗은 채 허공을 가리키고 있는 수리영의 절박한 표정.\n\nLOCATION (lock): The dining area of the rooftop home's living room, centered on the now-clean table under the home's nighttime lighting.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Tracking around the near side of the dining table settles at chest height into a close-medium, nearly frontal view justified by Suriyeong’s desperate attempt to prove what she saw. Her face holds the upper center while both extended hands enter the lower foreground above the clean tabletop, indicating two separate empty places before the camera continues laterally.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: empty dining tabletop beneath Suriyeong's hands in the lower-center of the frame, foreground; 수리영 in the middle-center of the frame, midground, points to two empty places on the table.\n- KEY BACKGROUND ELEMENTS: 식탁 (Clean; two claimed teacups absent) — The top plane angles toward the camera beneath her hands; used as Supports Suriyeong’s extended hands and exposes the two places where she says cups had been.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral interior ambience is shaped into moody low-key contrast with a restrained, tactile muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S14sh8__bgfirst_bg.png", "asset_id": "09b595f0-a89e-45ed-8cc7-0e638b43904f", "input_asset_ids": [ "308761ec-8143-4d37-b07f-4046d79cd4f2", "d1dd7f71-dbd5-43d7-bfbb-2c1fdf8ecfbd" ] }, "S14sh8": { "input_fingerprint": "f8b057cf87bd8c96", "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): 식탁 위로 두 손을 뻗은 채 허공을 가리키고 있는 수리영의 절박한 표정.\n\nLOCATION (lock): The dining area of the rooftop home's living room, centered on the now-clean table under the home's nighttime lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Tracking around the near side of the dining table settles at chest height into a close-medium, nearly frontal view justified by Suriyeong’s desperate attempt to prove what she saw. Her face holds the upper center while both extended hands enter the lower foreground above the clean tabletop, indicating two separate empty places before the camera continues laterally.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: empty dining tabletop beneath Suriyeong's hands in the lower-center of the frame, foreground; 수리영 in the middle-center of the frame, midground, points to two empty places on the table.\n- KEY BACKGROUND ELEMENTS: 식탁 (Clean; two claimed teacups absent) — The top plane angles toward the camera beneath her hands; used as Supports Suriyeong’s extended hands and exposes the two places where she says cups had been.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral interior ambience is shaped into moody low-key contrast with a restrained, tactile muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The dining table is now completely clean, with both tea cups—including the overturned one—missing.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 식탁 위로 두 손을 뻗은 채 허공을 가리키고 있는 수리영의 절박한 표정.\n\nLOCATION (lock): The dining area of the rooftop home's living room, centered on the now-clean table under the home's nighttime lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Tracking around the near side of the dining table settles at chest height into a close-medium, nearly frontal view justified by Suriyeong’s desperate attempt to prove what she saw. Her face holds the upper center while both extended hands enter the lower foreground above the clean tabletop, indicating two separate empty places before the camera continues laterally.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: empty dining tabletop beneath Suriyeong's hands in the lower-center of the frame, foreground; 수리영 in the middle-center of the frame, midground, points to two empty places on the table.\n- KEY BACKGROUND ELEMENTS: 식탁 (Clean; two claimed teacups absent) — The top plane angles toward the camera beneath her hands; used as Supports Suriyeong’s extended hands and exposes the two places where she says cups had been.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral interior ambience is shaped into moody low-key contrast with a restrained, tactile muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The dining table is now completely clean, with both tea cups—including the overturned one—missing.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 식탁 위로 두 손을 뻗은 채 허공을 가리키고 있는 수리영의 절박한 표정.\n\nLOCATION (lock): The dining area of the rooftop home's living room, centered on the now-clean table under the home's nighttime lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Tracking around the near side of the dining table settles at chest height into a close-medium, nearly frontal view justified by Suriyeong’s desperate attempt to prove what she saw. Her face holds the upper center while both extended hands enter the lower foreground above the clean tabletop, indicating two separate empty places before the camera continues laterally.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: empty dining tabletop beneath Suriyeong's hands in the lower-center of the frame, foreground; 수리영 in the middle-center of the frame, midground, points to two empty places on the table.\n- KEY BACKGROUND ELEMENTS: 식탁 (Clean; two claimed teacups absent) — The top plane angles toward the camera beneath her hands; used as Supports Suriyeong’s extended hands and exposes the two places where she says cups had been.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral interior ambience is shaped into moody low-key contrast with a restrained, tactile muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The dining table is now completely clean, with both tea cups—including the overturned one—missing.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S14sh8__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S14sh8.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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B07.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" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "배경에 원래의 식탁이 그대로 있는 상태에서 전경에 새로운 식탁을 창조해 낸 심각한 공간 구조 오류(하드 위반)가 있으며, 손으로 허공을 가리키라는 지시와 달리 손을 펴고만 있습니다." }, { "label": "B", "score": 10, "verdict_ko": "레퍼런스 장소의 정확한 식탁 위치에서 카메라 구도를 잡았으며, 테이블 위 빈 곳을 가리키는 두 손의 구체적인 동작과 절박한 표정 등 프레이밍 지시를 완벽하게 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "배경에 원래의 식탁이 그대로 있는 상태에서 전경에 새로운 식탁을 창조해 낸 심각한 공간 구조 오류(하드 위반)가 있으며, 손으로 허공을 가리키라는 지시와 달리 손을 펴고만 있습니다." }, { "label": "B", "score": 10, "verdict_ko": "레퍼런스 장소의 정확한 식탁 위치에서 카메라 구도를 잡았으며, 테이블 위 빈 곳을 가리키는 두 손의 구체적인 동작과 절박한 표정 등 프레이밍 지시를 완벽하게 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 클로즈 미디엄 샷 프레이밍, 양손으로 빈 식탁을 가리키는 동작, 그리고 절박한 표정을 참조된 공간 내에서 매우 정확하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "참조 이미지에 없는 흰색 테이블을 전경에 임의로 추가하여 공간 구성을 위반했으며, 양손으로 가리키는 동작 대신 손을 펴고 있어 지시문과 일치하지 않습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 클로즈 미디엄 샷 프레이밍, 양손으로 빈 식탁을 가리키는 동작, 그리고 절박한 표정을 참조된 공간 내에서 매우 정확하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "참조 이미지에 없는 흰색 테이블을 전경에 임의로 추가하여 공간 구성을 위반했으며, 양손으로 가리키는 동작 대신 손을 펴고 있어 지시문과 일치하지 않습니다." } ] }, "combined": { "totals": { "A": 5, "B": 18 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 5, "B": 18 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "배경에 원래의 식탁이 그대로 있는 상태에서 전경에 새로운 식탁을 창조해 낸 심각한 공간 구조 오류(하드 위반)가 있으며, 손으로 허공을 가리키라는 지시와 달리 손을 펴고만 있습니다." }, { "label": "B", "score": 10, "verdict_ko": "레퍼런스 장소의 정확한 식탁 위치에서 카메라 구도를 잡았으며, 테이블 위 빈 곳을 가리키는 두 손의 구체적인 동작과 절박한 표정 등 프레이밍 지시를 완벽하게 구현했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B07.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "화면 왼쪽(인물의 오른팔) 손의 엄지손가락이 바깥쪽을 향하고 있어 해부학적으로 잘못된 왼손의 형태를 띠고 있습니다.", "fix_en": "Replace the hand on the viewer's left with an anatomically correct right hand, placing the thumb on the inner side facing the other hand.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace the hand on the viewer's left with an anatomically correct right hand, placing the thumb on the inner side facing the other hand.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 프레이밍과 텅 빈 식탁 위로 두 손을 뻗어 빈자리를 가리키는 포즈를 자연스러운 손가락 해부학으로 정확히 구현함." }, { "label": "B", "score": 5, "verdict_ko": "프레이밍과 배경, 인물의 위치는 요구사항에 부합하나 뻗은 양손의 손가락 형태가 다소 부자연스러움." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 프레이밍과 텅 빈 식탁 위로 두 손을 뻗어 빈자리를 가리키는 포즈를 자연스러운 손가락 해부학으로 정확히 구현함." }, { "label": "B", "score": 5, "verdict_ko": "프레이밍과 배경, 인물의 위치는 요구사항에 부합하나 뻗은 양손의 손가락 형태가 다소 부자연스러움." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 샷 크기, 빈 식탁과 인물의 배치, 절박한 표정을 정확히 구현했으며 두 손의 해부학적 형태도 자연스럽습니다." }, { "label": "B", "score": 7, "verdict_ko": "요구된 구도와 감정 표현, 배경 조건은 잘 맞췄으나, 화면 왼쪽 손의 손가락 관절 형태가 다소 부자연스럽게 묘사되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 샷 크기, 빈 식탁과 인물의 배치, 절박한 표정을 정확히 구현했으며 두 손의 해부학적 형태도 자연스럽습니다." }, { "label": "A", "score": 7, "verdict_ko": "요구된 구도와 감정 표현, 배경 조건은 잘 맞췄으나, 화면 왼쪽 손의 손가락 관절 형태가 다소 부자연스럽게 묘사되었습니다." } ] }, "combined": { "totals": { "A": 14, "B": 13 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S14sh8__bgfirst_bg.png", "bg_asset_id": "09b595f0-a89e-45ed-8cc7-0e638b43904f", "bg_record_key": "S14sh8::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B07", "choice": "Candidate G", "confident": true, "reason_ko": "\"식탁 위로\"라는 지문과 함께 식탁이 있는 공간을 요구하며, 지정된 후보가 해당 위치를 포함하고 있으므로 현재 할당을 유지합니다.", "kept": "L04B07" } }, "S14sh11::variants": { "author_fp": "e58746bc522ef8f4", "author": { "variants": [ { "approach_ko": "얕은 심도로 두 여성의 맞물린 시선과 어깨를 붙든 손의 감정적 압박을 응축한다.", "prompt_en": "Use a gently compressed, intimate medium-shot perspective with very shallow depth, holding both faces and Hyesu’s hands in the same narrow plane of sharp focus while the room opening and table edge fall into soft, subdued separation. Shape the existing dim interior light into restrained facial modeling, with delicate highlights on eyes, fingertips, and fabric creases; keep contrast soft and the background visually quiet. Emphasize the locked eyeline and the physical reassurance of the grip as the emotional center." }, { "approach_ko": "더 깊은 초점과 열린 공간감을 사용해 두 사람의 긴장과 뒤편 방으로 이어질 시선의 방향을 함께 드러낸다.", "prompt_en": "Photograph the prescribed medium two-shot with a more open, spatial lens feel and moderately deep focus, preserving clear depth from the women to the receding room opening while keeping the partial table edge peripheral and unobtrusive. Let the composition breathe toward the room behind them, using layered planes and restrained negative space to suggest the next visual direction. Keep the dim ambient illumination broad and low-contrast, with natural falloff across faces, hands, clothing, and the surrounding interior rather than isolating the pair with a pronounced focus effect." } ] }, "reused": false }, "S14sh11": { "input_fingerprint": "b9a1056416fbaf96", "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): 수리영의 어깨를 양손으로 감싸 쥔 채 눈을 맞추고 있는 혜수의 상반신.\n\nLOCATION (lock): Inside the rooftop home's living area near the bedroom and dining table, with the dim nighttime interior lights on. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The forward dolly resolves at chest height into an upper-body two-shot, placing Hyesu slightly left and Suriyeong slightly right on a shared plane. Hyesu’s hands remain fully visible around Suriyeong’s shoulders, and their opposing three-quarter angles lock the eyeline without either woman addressing the camera.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 식탁 (Clean; claimed teacups absent) — Only the near corner and top edge remain at the frame boundary; used as A partial edge maintains continuity with the preceding search without competing with the two women; 수리영 방 안쪽 (Belongings remain inside) — The room opening recedes behind the pair; used as Provides the direction toward which the camera will pan after Hyesu steadies Suriyeong.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior light is kept subdued and low-contrast on the background while restrained highlights preserve Hyesu’s hands and the women’s locked gaze.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dining-area background, clean tabletop state, night lighting, and the younger woman's clothing and position near the table. Exclude her earlier pointing gesture and any absent cups, while adding only the older detective holding her shoulders.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The corpse, blood, red circle, and tea cups remain absent. Su-ri-young’s backpack and doll are present in her room but have been left in slightly altered positions.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 수리영의 어깨를 양손으로 감싸 쥔 채 눈을 맞추고 있는 혜수의 상반신.\n\nLOCATION (lock): Inside the rooftop home's living area near the bedroom and dining table, with the dim nighttime interior lights on. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The forward dolly resolves at chest height into an upper-body two-shot, placing Hyesu slightly left and Suriyeong slightly right on a shared plane. Hyesu’s hands remain fully visible around Suriyeong’s shoulders, and their opposing three-quarter angles lock the eyeline without either woman addressing the camera.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 식탁 (Clean; claimed teacups absent) — Only the near corner and top edge remain at the frame boundary; used as A partial edge maintains continuity with the preceding search without competing with the two women; 수리영 방 안쪽 (Belongings remain inside) — The room opening recedes behind the pair; used as Provides the direction toward which the camera will pan after Hyesu steadies Suriyeong.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior light is kept subdued and low-contrast on the background while restrained highlights preserve Hyesu’s hands and the women’s locked gaze.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dining-area background, clean tabletop state, night lighting, and the younger woman's clothing and position near the table. Exclude her earlier pointing gesture and any absent cups, while adding only the older detective holding her shoulders.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The corpse, blood, red circle, and tea cups remain absent. Su-ri-young’s backpack and doll are present in her room but have been left in slightly altered positions.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 gently compressed, intimate medium-shot perspective with very shallow depth, holding both faces and Hyesu’s hands in the same narrow plane of sharp focus while the room opening and table edge fall into soft, subdued separation. Shape the existing dim interior light into restrained facial modeling, with delicate highlights on eyes, fingertips, and fabric creases; keep contrast soft and the background visually quiet. Emphasize the locked eyeline and the physical reassurance of the grip as the emotional center.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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): 수리영의 어깨를 양손으로 감싸 쥔 채 눈을 맞추고 있는 혜수의 상반신.\n\nLOCATION (lock): Inside the rooftop home's living area near the bedroom and dining table, with the dim nighttime interior lights on. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: The forward dolly resolves at chest height into an upper-body two-shot, placing Hyesu slightly left and Suriyeong slightly right on a shared plane. Hyesu’s hands remain fully visible around Suriyeong’s shoulders, and their opposing three-quarter angles lock the eyeline without either woman addressing the camera.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: 식탁 (Clean; claimed teacups absent) — Only the near corner and top edge remain at the frame boundary; used as A partial edge maintains continuity with the preceding search without competing with the two women; 수리영 방 안쪽 (Belongings remain inside) — The room opening recedes behind the pair; used as Provides the direction toward which the camera will pan after Hyesu steadies Suriyeong.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient interior light is kept subdued and low-contrast on the background while restrained highlights preserve Hyesu’s hands and the women’s locked gaze.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dining-area background, clean tabletop state, night lighting, and the younger woman's clothing and position near the table. Exclude her earlier pointing gesture and any absent cups, while adding only the older detective holding her shoulders.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The corpse, blood, red circle, and tea cups remain absent. Su-ri-young’s backpack and doll are present in her room but have been left in slightly altered positions.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 prescribed medium two-shot with a more open, spatial lens feel and moderately deep focus, preserving clear depth from the women to the receding room opening while keeping the partial table edge peripheral and unobtrusive. Let the composition breathe toward the room behind them, using layered planes and restrained negative space to suggest the next visual direction. Keep the dim ambient illumination broad and low-contrast, with natural falloff across faces, hands, clothing, and the surrounding interior rather than isolating the pair with a pronounced focus 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": { "B": 7, "A": 5 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 상반신 투샷과 시선 교환을 잘 구현했으며, 특히 배경 방 안의 인형과 배낭을 정확히 묘사하여 디테일에서 앞섭니다." }, { "label": "A", "score": 5, "verdict_ko": "인물 간의 구도와 시선 처리는 양호하나, 프롬프트에서 요구한 배경 방 안의 소품(인형과 배낭)이 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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S14sh8_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "혜수의 양손이 수리영의 어깨를 모두 감싸 쥐고 완전히 보여야 한다는 지시와 달리 한 손만 보입니다.", "fix_en": "Add Hyesu's right hand visibly grasping Suriyeong's right shoulder.", "unfixable": false }, { "issue_ko": "두 인물이 3/4 각도(three-quarter angles)로 마주보아야 한다는 지시와 달리 완전한 측면(profile)으로 서 있습니다.", "fix_en": "Rotate both characters' bodies to a three-quarter angle facing towards each other.", "unfixable": true }, { "issue_ko": "식탁은 프레임 경계에 모서리와 윗부분만 일부 보여야 한다는 지시와 달리, 왼쪽 배경에 식탁 전체와 의자가 너무 많이 노출되었습니다.", "fix_en": "Remove the chairs and reduce the table's presence so only its near corner is visible at the frame boundary.", "unfixable": true }, { "issue_ko": "레퍼런스 이미지의 혜수는 경찰 모자를 쓰고 있으나, 생성된 이미지에서는 모자 없이 포니테일 머리를 하고 있습니다.", "fix_en": "Add the navy police cap to Hyesu's head and remove the ponytail to match the character reference.", "unfixable": false }, { "issue_ko": "화면 내에 어떠한 텍스트도 없어야 한다는 지시와 달리 혜수의 재킷 어깨와 가슴 부분에 깨진 글자와 기호가 있습니다.", "fix_en": "Erase all white text and symbols from Hyesu's jacket.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Reduce the dining table to only a narrow near-corner and top-edge sliver at the left frame boundary, filling the removed area with the existing surrounding interior while preserving both women unchanged.", "issue_ko": "식탁이 화면 왼쪽 하단에 넓고 전경 크게 드러나 의도된 ‘프레임 경계의 가까운 모서리와 상단 가장자리만 보이는’ 주변 요소보다 지나치게 두드러진다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add Hyesu's right hand visibly grasping Suriyeong's right shoulder.\n- Add the navy police cap to Hyesu's head and remove the ponytail to match the character reference.\n- Erase all white text and symbols from Hyesu's jacket.\n- Reduce the dining table to only a narrow near-corner and top-edge sliver at the left frame boundary, filling the removed area with the existing surrounding interior while preserving both women unchanged.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 구도와 인물의 동작을 잘 구현했으며, 레퍼런스의 경찰 모자를 포함하여 복장 고증이 더 충실합니다." }, { "label": "A", "score": 5, "verdict_ko": "지시된 샷의 구도와 동작은 대체로 일치하나, 혜수 캐릭터 레퍼런스에 있는 경찰 모자가 누락되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 구도와 인물의 동작을 잘 구현했으며, 레퍼런스의 경찰 모자를 포함하여 복장 고증이 더 충실합니다." }, { "label": "A", "score": 5, "verdict_ko": "지시된 샷의 구도와 동작은 대체로 일치하나, 혜수 캐릭터 레퍼런스에 있는 경찰 모자가 누락되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "샷 텍스트가 요구하는 상반신 투샷과 시선 교환을 정확히 연출했으며, 레퍼런스에 제시된 혜수의 경찰 모자 등 복장을 충실히 반영했습니다." }, { "label": "B", "score": 5, "verdict_ko": "구도와 배경 요소는 지시사항을 잘 따랐으나, 캐릭터 레퍼런스에 명시된 혜수의 경찰 모자가 누락되어 복장 일관성에서 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "샷 텍스트가 요구하는 상반신 투샷과 시선 교환을 정확히 연출했으며, 레퍼런스에 제시된 혜수의 경찰 모자 등 복장을 충실히 반영했습니다." }, { "label": "A", "score": 5, "verdict_ko": "구도와 배경 요소는 지시사항을 잘 따랐으나, 캐릭터 레퍼런스에 명시된 혜수의 경찰 모자가 누락되어 복장 일관성에서 감점되었습니다." } ] }, "combined": { "totals": { "A": 10, "B": 14 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S14sh8" } }, "groupbg::해변 정류장": { "input_fingerprint": "3d24800fc7bd5c25", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "해변 정류장", "tags": [ "S15sh1", "S15sh5" ] }, "context_sig": "e5b0d9eae46686de" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: An outdoor bench at a coastal bus stop, facing the dark sea at night.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안가 버스정류장: 바다 풍경이 배경으로 펼쳐지는 단출한 버스 승강장이다. (특징: 투명한 바람막이 구조물과 벤치; 정류장 표지판; 바다가 보이는 뷰)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#15. 해안가. 버스정류장 부근 / N\n- 바다를 바라보며 멍한 채로 앉아있는 수리영\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: An outdoor bench at a coastal bus stop, facing the dark sea at night.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n해안가 버스정류장: 바다 풍경이 배경으로 펼쳐지는 단출한 버스 승강장이다. (특징: 투명한 바람막이 구조물과 벤치; 정류장 표지판; 바다가 보이는 뷰)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#15. 해안가. 버스정류장 부근 / N\n- 바다를 바라보며 멍한 채로 앉아있는 수리영\n\nTIME OF DAY (lock): night.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해변_정류장_dc431d.png", "asset_id": "f321b9df-2406-4ef7-ad8e-d5e3f2b17413", "input_asset_ids": [ "fb143afb-7c01-43bb-95f6-a873985ac297" ], "origin_tag": "S15sh1", "place_text": "An outdoor bench at a coastal bus stop, facing the dark sea at night.", "origin_inputs": { "place_text": "An outdoor bench at a coastal bus stop, facing the dark sea at night.", "time_of_day_en": "night", "conti_asset_id": "fb143afb-7c01-43bb-95f6-a873985ac297" } }, "S15sh1::bgfirst_bg": { "input_fingerprint": "69edaf93d13724ef", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 밤바다를 배경으로 버스정류장 벤치에 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bench at a coastal bus stop, facing the dark sea at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At Suriyeong’s seated eye level and roughly two and a half meters from the side of the bus-stop bench, the static camera looks across the bench toward the night sea. Suriyeong’s curled profile sits off-center with surrounding bench space and the sea left visible, emphasizing her isolation before the later approach.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (Occupied by Suriyeong) — Its long side crosses the frame toward the sea; used as Supports Suriyeong and extends empty space around her body; 밤바다 (Visible at night); used as Forms the distant background along Suriyeong’s outward gaze; 버스정류장 구조 (Visible) — A partial side faces the camera beside the bench; used as Locates the bench within the coastal bus-stop area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the coastal night is graded in muted neutrals with low-key contrast and subtly rough texture.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 밤바다를 배경으로 버스정류장 벤치에 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bench at a coastal bus stop, facing the dark sea at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At Suriyeong’s seated eye level and roughly two and a half meters from the side of the bus-stop bench, the static camera looks across the bench toward the night sea. Suriyeong’s curled profile sits off-center with surrounding bench space and the sea left visible, emphasizing her isolation before the later approach.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (Occupied by Suriyeong) — Its long side crosses the frame toward the sea; used as Supports Suriyeong and extends empty space around her body; 밤바다 (Visible at night); used as Forms the distant background along Suriyeong’s outward gaze; 버스정류장 구조 (Visible) — A partial side faces the camera beside the bench; used as Locates the bench within the coastal bus-stop area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the coastal night is graded in muted neutrals with low-key contrast and subtly rough texture.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S15sh1__bgfirst_bg.png", "asset_id": "9f75f7c0-4a9a-42a7-83e7-d5b390d06e57", "input_asset_ids": [ "fb143afb-7c01-43bb-95f6-a873985ac297", "f321b9df-2406-4ef7-ad8e-d5e3f2b17413" ] }, "S15sh1": { "input_fingerprint": "d86538c82d414530", "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): 밤바다를 배경으로 버스정류장 벤치에 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bench at a coastal bus stop, facing the dark sea at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At Suriyeong’s seated eye level and roughly two and a half meters from the side of the bus-stop bench, the static camera looks across the bench toward the night sea. Suriyeong’s curled profile sits off-center with surrounding bench space and the sea left visible, emphasizing her isolation before the later approach.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (Occupied by Suriyeong) — Its long side crosses the frame toward the sea; used as Supports Suriyeong and extends empty space around her body; 밤바다 (Visible at night); used as Forms the distant background along Suriyeong’s outward gaze; 버스정류장 구조 (Visible) — A partial side faces the camera beside the bench; used as Locates the bench within the coastal bus-stop area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the coastal night is graded in muted neutrals with low-key contrast and subtly rough texture.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 밤바다를 배경으로 버스정류장 벤치에 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bench at a coastal bus stop, facing the dark sea at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At Suriyeong’s seated eye level and roughly two and a half meters from the side of the bus-stop bench, the static camera looks across the bench toward the night sea. Suriyeong’s curled profile sits off-center with surrounding bench space and the sea left visible, emphasizing her isolation before the later approach.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (Occupied by Suriyeong) — Its long side crosses the frame toward the sea; used as Supports Suriyeong and extends empty space around her body; 밤바다 (Visible at night); used as Forms the distant background along Suriyeong’s outward gaze; 버스정류장 구조 (Visible) — A partial side faces the camera beside the bench; used as Locates the bench within the coastal bus-stop area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the coastal night is graded in muted neutrals with low-key contrast and subtly rough texture.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 밤바다를 배경으로 버스정류장 벤치에 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): An outdoor bench at a coastal bus stop, facing the dark sea at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At Suriyeong’s seated eye level and roughly two and a half meters from the side of the bus-stop bench, the static camera looks across the bench toward the night sea. Suriyeong’s curled profile sits off-center with surrounding bench space and the sea left visible, emphasizing her isolation before the later approach.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 버스정류장 벤치 (Occupied by Suriyeong) — Its long side crosses the frame toward the sea; used as Supports Suriyeong and extends empty space around her body; 밤바다 (Visible at night); used as Forms the distant background along Suriyeong’s outward gaze; 버스정류장 구조 (Visible) — A partial side faces the camera beside the bench; used as Locates the bench within the coastal bus-stop area.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural ambient light appropriate to the coastal night is graded in muted neutrals with low-key contrast and subtly rough texture.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S15sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S15sh1.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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해변_정류장_dc431d.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" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 웅크린 측면 자세와 와이드 샷 프레이밍을 정확히 구현했으며, 벤치와의 물리적 접촉과 캐릭터 외모가 자연스럽습니다." }, { "label": "A", "score": 6, "verdict_ko": "구도와 장소는 지침을 잘 따랐으나, 캐릭터의 얼굴 묘사가 레퍼런스와 다소 차이가 있고 발이 놓인 형태가 어색합니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 웅크린 측면 자세와 와이드 샷 프레이밍을 정확히 구현했으며, 벤치와의 물리적 접촉과 캐릭터 외모가 자연스럽습니다." }, { "label": "A", "score": 6, "verdict_ko": "구도와 장소는 지침을 잘 따랐으나, 캐릭터의 얼굴 묘사가 레퍼런스와 다소 차이가 있고 발이 놓인 형태가 어색합니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 앵글과 스케일을 정확히 구현했으며, 벤치에 '웅크리고 앉아 있는' 포즈를 B보다 더 자연스럽고 완벽하게 표현했습니다." }, { "label": "B", "score": 7, "verdict_ko": "프레이밍과 배경 구현은 우수하나, 캐릭터의 상체가 꼿꼿해 '웅크린' 느낌이 상대적으로 덜하며 손 부분의 묘사가 다소 아쉽습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 앵글과 스케일을 정확히 구현했으며, 벤치에 '웅크리고 앉아 있는' 포즈를 B보다 더 자연스럽고 완벽하게 표현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "프레이밍과 배경 구현은 우수하나, 캐릭터의 상체가 꼿꼿해 '웅크린' 느낌이 상대적으로 덜하며 손 부분의 묘사가 다소 아쉽습니다." } ] }, "combined": { "totals": { "A": 13, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 13, "B": 16 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 웅크린 측면 자세와 와이드 샷 프레이밍을 정확히 구현했으며, 벤치와의 물리적 접촉과 캐릭터 외모가 자연스럽습니다." }, { "label": "A", "score": 6, "verdict_ko": "구도와 장소는 지침을 잘 따랐으나, 캐릭터의 얼굴 묘사가 레퍼런스와 다소 차이가 있고 발이 놓인 형태가 어색합니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해변_정류장_dc431d.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [] }, "composition_critique": { "issues": [] }, "fix_skipped": true, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S15sh1__bgfirst_bg.png", "bg_asset_id": "9f75f7c0-4a9a-42a7-83e7-d5b390d06e57", "bg_record_key": "S15sh1::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "해변 정류장", "groupbg_asset_id": "f321b9df-2406-4ef7-ad8e-d5e3f2b17413" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S15sh5::variants": { "author_fp": "1a28aa4f14514c21", "author": { "variants": [ { "approach_ko": "혜수를 전경의 타이트한 상반신 클로즈업으로 잡고, 멀어지는 수리영을 얕은 심도 속 배경 동작으로 분리한다.", "prompt_en": "Frame a tight upper-body close-up from near eye level, placing Hyesu prominently in the foreground while Suriyeong recedes behind her along the frame’s depth. Use a compressed lens feel and shallow focus: Hyesu’s cupped hands, open mouth, and urgent facial performance are crisp, while Suriyeong’s raised-foot running stride remains clearly readable but softly separated against the dim coast. Let cool coastal night light skim faces and fabric with restrained contrast and natural falloff." }, { "approach_ko": "낮고 약간 비껴선 넓은 시점에서 두 사람을 대각선 깊이로 연결하고, 수리영의 달리는 순간까지 선명하게 살린다.", "prompt_en": "Photograph from a slightly low, oblique viewpoint with a broader lens feel, keeping Hyesu’s upper body large near one side of the frame and directing the composition diagonally toward Suriyeong in the background. Hold deeper focus so the calling gesture and the airborne-foot running pose register together as one continuous moment, with the bus-stop structure and coastal space providing layered depth. Shape the muted night ambience into gentle edge separation rather than dramatic illumination, preserving realistic darkness and physical surface detail." } ] }, "reused": false }, "S15sh5": { "input_fingerprint": "04e615f911286951", "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): 한쪽 발이 공중에 뜬 달리기 자세인 수리영의 뒷모습을 배경으로, 두 손을 입가에 모으고 입을 크게 벌린 혜수의 상반신.\n\nLOCATION (lock): Beside the outdoor coastal bus stop, with the departing path and night sea behind it. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural coastal night ambience remains low-key and muted, with enough separation to preserve Suriyeong’s running silhouette behind Hyesu.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 night sea, bus-stop structure, bench, and cool coastal lighting from the reference. Preserve both women's clothing, but exclude the younger woman's seated pose and frame her running away while the older woman calls after her.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 한쪽 발이 공중에 뜬 달리기 자세인 수리영의 뒷모습을 배경으로, 두 손을 입가에 모으고 입을 크게 벌린 혜수의 상반신.\n\nLOCATION (lock): Beside the outdoor coastal bus stop, with the departing path and night sea behind it. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural coastal night ambience remains low-key and muted, with enough separation to preserve Suriyeong’s running silhouette behind Hyesu.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 night sea, bus-stop structure, bench, and cool coastal lighting from the reference. Preserve both women's clothing, but exclude the younger woman's seated pose and frame her running away while the older woman calls after her.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nFrame a tight upper-body close-up from near eye level, placing Hyesu prominently in the foreground while Suriyeong recedes behind her along the frame’s depth. Use a compressed lens feel and shallow focus: Hyesu’s cupped hands, open mouth, and urgent facial performance are crisp, while Suriyeong’s raised-foot running stride remains clearly readable but softly separated against the dim coast. Let cool coastal night light skim faces and fabric with restrained contrast and natural falloff.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 한쪽 발이 공중에 뜬 달리기 자세인 수리영의 뒷모습을 배경으로, 두 손을 입가에 모으고 입을 크게 벌린 혜수의 상반신.\n\nLOCATION (lock): Beside the outdoor coastal bus stop, with the departing path and night sea behind it. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural coastal night ambience remains low-key and muted, with enough separation to preserve Suriyeong’s running silhouette behind Hyesu.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 night sea, bus-stop structure, bench, and cool coastal lighting from the reference. Preserve both women's clothing, but exclude the younger woman's seated pose and frame her running away while the older woman calls after her.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 slightly low, oblique viewpoint with a broader lens feel, keeping Hyesu’s upper body large near one side of the frame and directing the composition diagonally toward Suriyeong in the background. Hold deeper focus so the calling gesture and the airborne-foot running pose register together as one continuous moment, with the bus-stop structure and coastal space providing layered depth. Shape the muted night ambience into gentle edge separation rather than dramatic illumination, preserving realistic darkness and physical surface 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": { "B": 7, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "수리영의 역동적인 달리기 자세와 혜수의 시선 방향이 정확히 일치하여 상황의 맥락을 잘 살렸으나, 혜수의 자켓에 경찰 마크와 텍스트가 누락되었습니다." }, { "label": "A", "score": 3, "verdict_ko": "수리영의 달리기 자세가 팔을 내린 채 굳어 있어 부자연스러우며, 혜수가 수리영이 있는 곳과 완전히 다른 방향을 향해 소리치고 있어 상호작용이 어색합니다." } ], "refs": [ { "label": "STORYBOARD SKETCH — a bare thin-line staging guide, a REFERENCE ONLY: take from it ONLY the camera framing, figure placement, heading and size/depth order. It carries ZERO visual style — every texture, material, light and all realism come from the text and the photographic references. Never let any line-drawing quality leak into the output.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/lane_sketch_S15sh5.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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S15sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "혜수(앞쪽 인물)가 레퍼런스 이미지에 있는 경찰 모자를 쓰고 있지 않습니다.", "fix_en": "Add the dark navy police cap with the front emblem to Hyesu's head.", "unfixable": false }, { "issue_ko": "혜수의 재킷이 레퍼런스의 경찰 제복이 아닌 일반 남색 점퍼로 묘사되어 견장과 소매 패치 등의 디테일이 누락되었습니다.", "fix_en": "Modify Hyesu's jacket to match the police uniform reference by adding shoulder epaulettes and sleeve patches.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add the dark navy police cap with the front emblem to Hyesu's head.\n- Modify Hyesu's jacket to match the police uniform reference by adding shoulder epaulettes and sleeve patches.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "샷 텍스트의 구도와 자세를 완벽히 구현했으며, 혜수의 경찰 제복과 모자 등 의상 디테일을 레퍼런스에 맞게 정확히 유지했습니다." }, { "label": "A", "score": 5, "verdict_ko": "구도와 인물의 자세는 지시를 잘 따랐으나, 혜수의 의상에서 레퍼런스에 있는 경찰 모자와 마크가 누락되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "샷 텍스트의 구도와 자세를 완벽히 구현했으며, 혜수의 경찰 제복과 모자 등 의상 디테일을 레퍼런스에 맞게 정확히 유지했습니다." }, { "label": "A", "score": 5, "verdict_ko": "구도와 인물의 자세는 지시를 잘 따랐으나, 혜수의 의상에서 레퍼런스에 있는 경찰 모자와 마크가 누락되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "혜수의 모자와 경찰 제복 디테일을 레퍼런스와 가깝게 유지했으며, 두 사람의 지시된 포즈와 카메라 구도를 정확히 구현함." }, { "label": "B", "score": 6, "verdict_ko": "지시된 포즈와 배경 연출은 훌륭하나, 혜수의 경찰 모자가 누락되고 제복의 패치 등 필수 복장 디테일이 사라짐." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "혜수의 모자와 경찰 제복 디테일을 레퍼런스와 가깝게 유지했으며, 두 사람의 지시된 포즈와 카메라 구도를 정확히 구현함." }, { "label": "A", "score": 6, "verdict_ko": "지시된 포즈와 배경 연출은 훌륭하나, 혜수의 경찰 모자가 누락되고 제복의 패치 등 필수 복장 디테일이 사라짐." } ] }, "combined": { "totals": { "A": 11, "B": 17 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "lane(map_marker): 스케치+prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S15sh1" } }, "S16sh1::bgfirst_bg": { "input_fingerprint": "75a104009f1204e9", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 불 꺼진 경찰서 복도에서 환한 사무실 문을 반쯤 열고 한 발을 안으로 내디딘 채 귀에 휴대폰을 대고 있는 혜수의 전신.\n\nLOCATION (lock): In the dark police-station corridor at the threshold of a brightly lit duty office at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level several steps down the unlit corridor, the static camera faces the half-open office door and frames Hyesu’s full body as she crosses the threshold. She is positioned just off-center with one foot already inside, the phone pressed to her ear, and the brighter office visible beyond her as the destination of the next dolly move.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground, moves toward police office interior; police office interior in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 경찰서 복도 (Lights off) — The corridor recedes directly toward the office doorway; used as Provides the dark foreground approach and outlines Hyesu’s entrance path; 사무실 문 (Half-open) — The corridor-facing side and partial opening are visible; used as Frames Hyesu at the instant she steps into the office; 경찰서 사무실 (Brightly lit) — The interior is visible through the half-open doorway; used as Creates the brighter destination behind Hyesu.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The bright office against the unlit corridor creates stark low-key threshold contrast within a restrained muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 불 꺼진 경찰서 복도에서 환한 사무실 문을 반쯤 열고 한 발을 안으로 내디딘 채 귀에 휴대폰을 대고 있는 혜수의 전신.\n\nLOCATION (lock): In the dark police-station corridor at the threshold of a brightly lit duty office at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level several steps down the unlit corridor, the static camera faces the half-open office door and frames Hyesu’s full body as she crosses the threshold. She is positioned just off-center with one foot already inside, the phone pressed to her ear, and the brighter office visible beyond her as the destination of the next dolly move.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground, moves toward police office interior; police office interior in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 경찰서 복도 (Lights off) — The corridor recedes directly toward the office doorway; used as Provides the dark foreground approach and outlines Hyesu’s entrance path; 사무실 문 (Half-open) — The corridor-facing side and partial opening are visible; used as Frames Hyesu at the instant she steps into the office; 경찰서 사무실 (Brightly lit) — The interior is visible through the half-open doorway; used as Creates the brighter destination behind Hyesu.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The bright office against the unlit corridor creates stark low-key threshold contrast within a restrained muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S16sh1__bgfirst_bg.png", "asset_id": "50ea5ec7-a6b1-47e7-bb7b-0416d511a462", "input_asset_ids": [ "deb80e81-d0f2-4c2b-b470-8645061cefbf", "03803359-3162-4570-9f5e-3d2729760ba7" ] }, "S16sh1": { "input_fingerprint": "73348c5277a6bcc0", "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): 불 꺼진 경찰서 복도에서 환한 사무실 문을 반쯤 열고 한 발을 안으로 내디딘 채 귀에 휴대폰을 대고 있는 혜수의 전신.\n\nLOCATION (lock): In the dark police-station corridor at the threshold of a brightly lit duty office at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level several steps down the unlit corridor, the static camera faces the half-open office door and frames Hyesu’s full body as she crosses the threshold. She is positioned just off-center with one foot already inside, the phone pressed to her ear, and the brighter office visible beyond her as the destination of the next dolly move.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground, moves toward police office interior; police office interior in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 경찰서 복도 (Lights off) — The corridor recedes directly toward the office doorway; used as Provides the dark foreground approach and outlines Hyesu’s entrance path; 사무실 문 (Half-open) — The corridor-facing side and partial opening are visible; used as Frames Hyesu at the instant she steps into the office; 경찰서 사무실 (Brightly lit) — The interior is visible through the half-open doorway; used as Creates the brighter destination behind Hyesu.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The bright office against the unlit corridor creates stark low-key threshold contrast within a restrained muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo continues carrying and using her phone to call Min-sook as she enters the office.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 불 꺼진 경찰서 복도에서 환한 사무실 문을 반쯤 열고 한 발을 안으로 내디딘 채 귀에 휴대폰을 대고 있는 혜수의 전신.\n\nLOCATION (lock): In the dark police-station corridor at the threshold of a brightly lit duty office at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level several steps down the unlit corridor, the static camera faces the half-open office door and frames Hyesu’s full body as she crosses the threshold. She is positioned just off-center with one foot already inside, the phone pressed to her ear, and the brighter office visible beyond her as the destination of the next dolly move.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground, moves toward police office interior; police office interior in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 경찰서 복도 (Lights off) — The corridor recedes directly toward the office doorway; used as Provides the dark foreground approach and outlines Hyesu’s entrance path; 사무실 문 (Half-open) — The corridor-facing side and partial opening are visible; used as Frames Hyesu at the instant she steps into the office; 경찰서 사무실 (Brightly lit) — The interior is visible through the half-open doorway; used as Creates the brighter destination behind Hyesu.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The bright office against the unlit corridor creates stark low-key threshold contrast within a restrained muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo continues carrying and using her phone to call Min-sook as she enters the office.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 불 꺼진 경찰서 복도에서 환한 사무실 문을 반쯤 열고 한 발을 안으로 내디딘 채 귀에 휴대폰을 대고 있는 혜수의 전신.\n\nLOCATION (lock): In the dark police-station corridor at the threshold of a brightly lit duty office at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From eye level several steps down the unlit corridor, the static camera faces the half-open office door and frames Hyesu’s full body as she crosses the threshold. She is positioned just off-center with one foot already inside, the phone pressed to her ear, and the brighter office visible beyond her as the destination of the next dolly move.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 혜수 in the middle-center of the frame, midground, moves toward police office interior; police office interior in the upper-center of the frame, background.\n- KEY BACKGROUND ELEMENTS: 경찰서 복도 (Lights off) — The corridor recedes directly toward the office doorway; used as Provides the dark foreground approach and outlines Hyesu’s entrance path; 사무실 문 (Half-open) — The corridor-facing side and partial opening are visible; used as Frames Hyesu at the instant she steps into the office; 경찰서 사무실 (Brightly lit) — The interior is visible through the half-open doorway; used as Creates the brighter destination behind Hyesu.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The bright office against the unlit corridor creates stark low-key threshold contrast within a restrained muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo continues carrying and using her phone to call Min-sook as she enters the office.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S16sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S16sh1.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L13B03.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 대로 사무실 문턱에 걸쳐 서서 명암 대비를 잘 보여주나, 실내로 향해야 하는 방향성이 어긋나 카메라를 향해 걷고 있습니다." }, { "label": "A", "score": 4, "verdict_ko": "문턱에 한 발을 내디딘 상태라는 지시와 달리 이미 완전히 밝은 사무실 내부로 들어가 있어 프롬프트의 핵심 공간 연출을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 대로 사무실 문턱에 걸쳐 서서 명암 대비를 잘 보여주나, 실내로 향해야 하는 방향성이 어긋나 카메라를 향해 걷고 있습니다." }, { "label": "A", "score": 4, "verdict_ko": "문턱에 한 발을 내디딘 상태라는 지시와 달리 이미 완전히 밝은 사무실 내부로 들어가 있어 프롬프트의 핵심 공간 연출을 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "사무실로 진입하는 방향이 반대로 묘사되었으나, 프롬프트의 지시대로 캐릭터가 문턱에 걸쳐 있고 밝은 사무실이 캐릭터 뒤쪽 배경으로 오도록 레이아웃을 더 정확하게 구성했습니다." }, { "label": "A", "score": 4, "verdict_ko": "레퍼런스 이미지의 카메라 구도를 그대로 복사하여 사무실이 캐릭터 뒤가 아닌 측면에 위치하며, 문턱을 넘는 대신 복도에 완전히 나와 있어 공간 배치와 행동 지시를 어겼습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "사무실로 진입하는 방향이 반대로 묘사되었으나, 프롬프트의 지시대로 캐릭터가 문턱에 걸쳐 있고 밝은 사무실이 캐릭터 뒤쪽 배경으로 오도록 레이아웃을 더 정확하게 구성했습니다." }, { "label": "B", "score": 4, "verdict_ko": "레퍼런스 이미지의 카메라 구도를 그대로 복사하여 사무실이 캐릭터 뒤가 아닌 측면에 위치하며, 문턱을 넘는 대신 복도에 완전히 나와 있어 공간 배치와 행동 지시를 어겼습니다." } ] }, "combined": { "totals": { "A": 11, "B": 11 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 11 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 대로 사무실 문턱에 걸쳐 서서 명암 대비를 잘 보여주나, 실내로 향해야 하는 방향성이 어긋나 카메라를 향해 걷고 있습니다." }, { "label": "A", "score": 4, "verdict_ko": "문턱에 한 발을 내디딘 상태라는 지시와 달리 이미 완전히 밝은 사무실 내부로 들어가 있어 프롬프트의 핵심 공간 연출을 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L13B03.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "'한 발을 안으로 내디딘 채'라는 지문 및 문지방에 서 있는 레이아웃 스케치와 달리, 혜수의 두 발이 모두 사무실 안쪽에 위치해 있습니다.", "fix_en": "Move Hyesu backward onto the threshold so her left foot is outside in the corridor.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "issue_ko": "혜수가 레이아웃 스케치보다 화면에서 작고 너무 왼쪽에 배치되어 있다.", "fix_en": "Enlarge Hyesu slightly and shift her rightward within the doorway to match the layout sketch’s screen size and position.", "unfixable": false }, { "issue_ko": "혜수의 앞발과 몸의 진행 방향이 사무실 안쪽이 아니라 복도 쪽으로 나오는 동작으로 보인다.", "fix_en": "Adjust Hyesu’s stride so her leading foot is inside the office and her body is clearly moving deeper into the office.", "unfixable": false }, { "issue_ko": "기준 배경보다 출입구가 화면 왼쪽으로 치우치고 좁아졌으며 오른쪽 벽의 화면 점유가 커져, 고정된 카메라 프레이밍과 원근이 일치하지 않는다.", "fix_en": "Minimize the framing discrepancy with a slight right-edge crop so the doorway sits closer to its reference screen position.", "unfixable": true } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Move Hyesu backward onto the threshold so her left foot is outside in the corridor.\n- Enlarge Hyesu slightly and shift her rightward within the doorway to match the layout sketch’s screen size and position.\n- Adjust Hyesu’s stride so her leading foot is inside the office and her body is clearly moving deeper into the office.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "레이아웃 스케치에 제시된 정확한 자세(앞으로 뻗은 오른다리)를 따랐으며, 인물 레퍼런스의 이목구비와 의상의 텍스트 디테일까지 훌륭하게 재현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "스케치와 다리의 앞뒤 걷는 위치가 반대로 묘사되었으며, 경찰복의 텍스트가 뭉개지고 인물의 얼굴이 레퍼런스와 다소 차이를 보입니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "레이아웃 스케치에 제시된 정확한 자세(앞으로 뻗은 오른다리)를 따랐으며, 인물 레퍼런스의 이목구비와 의상의 텍스트 디테일까지 훌륭하게 재현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "스케치와 다리의 앞뒤 걷는 위치가 반대로 묘사되었으며, 경찰복의 텍스트가 뭉개지고 인물의 얼굴이 레퍼런스와 다소 차이를 보입니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "스케치의 구도를 정확히 따랐으며, 레퍼런스 이미지의 경찰 제복 디테일(가슴의 '경찰 POLICE' 로고)을 충실하게 구현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "전체적인 구도와 배경 유지는 훌륭하나, 제복 가슴의 로고가 레퍼런스와 다르게 뭉개졌고 내린 왼손의 손가락 형태가 다소 어색합니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "스케치의 구도를 정확히 따랐으며, 레퍼런스 이미지의 경찰 제복 디테일(가슴의 '경찰 POLICE' 로고)을 충실하게 구현했습니다." }, { "label": "B", "score": 7, "verdict_ko": "전체적인 구도와 배경 유지는 훌륭하나, 제복 가슴의 로고가 레퍼런스와 다르게 뭉개졌고 내린 왼손의 손가락 형태가 다소 어색합니다." } ] }, "combined": { "totals": { "A": 18, "B": 13 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S16sh1__bgfirst_bg.png", "bg_asset_id": "50ea5ec7-a6b1-47e7-bb7b-0416d511a462", "bg_record_key": "S16sh1::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L13B01", "B": "L13B02", "C": "L13B03", "D": "L13B04" }, "assigned": "L13B03", "choice": "Candidate C", "confident": true, "reason_ko": "\"불 꺼진 경찰서 복도에서 환한 사무실 문을 반쯤 열고\"라는 묘사에 정확히 부합하게, 어두운 복도 시점에서 밝게 불이 켜진 사무실 내부가 열린 문을 통해 보이는 구도를 담고 있습니다.", "kept": "L13B03" } }, "S16sh3::bgfirst_bg": { "input_fingerprint": "fa67ac8241888e94", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 휴대폰 화면을 켠 채 김 형사 눈앞으로 들이민 혜수의 굳은 손 클로즈업.\n\nLOCATION (lock): Inside the lit police duty office, beside the detective's workstation at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the near edge of the desk and hand height, the dolly-in ends on Hyesu’s rigid hand extending the active phone toward Kim’s off-frame position. The phone is angled so its displayed saved number is readable but occupies less than a third of the frame, while Hyesu’s hand and forearm provide scale and her face remains excluded.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 혜수의 휴대폰 화면 (On) — The active front screen faces both the camera and Kim’s position, displaying 민숙의 저장된 번호; used as Primary focal object held in front of Kim’s off-frame position; 책상 (In use) — The near edge crosses beneath the hand, with the top plane partially visible; used as Provides a grounded lower edge beneath Hyesu’s extended forearm.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Existing office illumination is controlled into subdued contrast, keeping the active screen readable without allowing it to become the scene’s sole light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 휴대폰 화면을 켠 채 김 형사 눈앞으로 들이민 혜수의 굳은 손 클로즈업.\n\nLOCATION (lock): Inside the lit police duty office, beside the detective's workstation at night.\n\nTIME OF DAY (lock): night.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the near edge of the desk and hand height, the dolly-in ends on Hyesu’s rigid hand extending the active phone toward Kim’s off-frame position. The phone is angled so its displayed saved number is readable but occupies less than a third of the frame, while Hyesu’s hand and forearm provide scale and her face remains excluded.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 혜수의 휴대폰 화면 (On) — The active front screen faces both the camera and Kim’s position, displaying 민숙의 저장된 번호; used as Primary focal object held in front of Kim’s off-frame position; 책상 (In use) — The near edge crosses beneath the hand, with the top plane partially visible; used as Provides a grounded lower edge beneath Hyesu’s extended forearm.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Existing office illumination is controlled into subdued contrast, keeping the active screen readable without allowing it to become the scene’s sole light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S16sh3__bgfirst_bg.png", "asset_id": "3f8030e6-ea5e-44d9-8d64-1f123ebfdfba", "input_asset_ids": [ "2102be36-643d-4639-a97b-9f0a63467b95", "447f50b0-1b71-4a75-81ba-f72d0a1b8533" ] }, "S16sh3": { "input_fingerprint": "c191045ffc5c59fd", "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): 휴대폰 화면을 켠 채 김 형사 눈앞으로 들이민 혜수의 굳은 손 클로즈업.\n\nLOCATION (lock): Inside the lit police duty office, beside the detective's workstation at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the near edge of the desk and hand height, the dolly-in ends on Hyesu’s rigid hand extending the active phone toward Kim’s off-frame position. The phone is angled so its displayed saved number is readable but occupies less than a third of the frame, while Hyesu’s hand and forearm provide scale and her face remains excluded.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 혜수의 휴대폰 화면 (On) — The active front screen faces both the camera and Kim’s position, displaying 민숙의 저장된 번호; used as Primary focal object held in front of Kim’s off-frame position; 책상 (In use) — The near edge crosses beneath the hand, with the top plane partially visible; used as Provides a grounded lower edge beneath Hyesu’s extended forearm.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Existing office illumination is controlled into subdued contrast, keeping the active screen readable without allowing it to become the scene’s sole light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo’s phone remains in her hand with Min-sook’s saved number displayed for Detective Kim.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일); 김 형사 (한국인 30대 남성, 평범한 성인 얼굴, 짙은 검은색 머리카락, 짧고 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 휴대폰 화면을 켠 채 김 형사 눈앞으로 들이민 혜수의 굳은 손 클로즈업.\n\nLOCATION (lock): Inside the lit police duty office, beside the detective's workstation at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the near edge of the desk and hand height, the dolly-in ends on Hyesu’s rigid hand extending the active phone toward Kim’s off-frame position. The phone is angled so its displayed saved number is readable but occupies less than a third of the frame, while Hyesu’s hand and forearm provide scale and her face remains excluded.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 혜수의 휴대폰 화면 (On) — The active front screen faces both the camera and Kim’s position, displaying 민숙의 저장된 번호; used as Primary focal object held in front of Kim’s off-frame position; 책상 (In use) — The near edge crosses beneath the hand, with the top plane partially visible; used as Provides a grounded lower edge beneath Hyesu’s extended forearm.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Existing office illumination is controlled into subdued contrast, keeping the active screen readable without allowing it to become the scene’s sole light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo’s phone remains in her hand with Min-sook’s saved number displayed for Detective Kim.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일); 김 형사 (한국인 30대 남성, 평범한 성인 얼굴, 짙은 검은색 머리카락, 짧고 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): night.\n\nSHOT TEXT (authoritative, Korean): 휴대폰 화면을 켠 채 김 형사 눈앞으로 들이민 혜수의 굳은 손 클로즈업.\n\nLOCATION (lock): Inside the lit police duty office, beside the detective's workstation at night. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the near edge of the desk and hand height, the dolly-in ends on Hyesu’s rigid hand extending the active phone toward Kim’s off-frame position. The phone is angled so its displayed saved number is readable but occupies less than a third of the frame, while Hyesu’s hand and forearm provide scale and her face remains excluded.\n- FRAMING SCALE: insert close-up on a detail\n- KEY BACKGROUND ELEMENTS: 혜수의 휴대폰 화면 (On) — The active front screen faces both the camera and Kim’s position, displaying 민숙의 저장된 번호; used as Primary focal object held in front of Kim’s off-frame position; 책상 (In use) — The near edge crosses beneath the hand, with the top plane partially visible; used as Provides a grounded lower edge beneath Hyesu’s extended forearm.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Existing office illumination is controlled into subdued contrast, keeping the active screen readable without allowing it to become the scene’s sole light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Hye-soo’s phone remains in her hand with Min-sook’s saved number displayed for Detective Kim.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일); 김 형사 (한국인 30대 남성, 평범한 성인 얼굴, 짙은 검은색 머리카락, 짧고 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S16sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S16sh3.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L13B04.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "요구된 인서트 클로즈업 앵글과 혜수의 팔, 책상의 배치 및 '민숙'이 적힌 휴대폰 화면을 지시사항에 맞게 사실적으로 잘 구현했습니다." }, { "label": "A", "score": 0, "verdict_ko": "인물의 머리가 없는 상태로 경찰 제복만 렌더링된 물리적으로 불가능한 해부학적 오류(하드 위반)가 발생하여 실격입니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "요구된 인서트 클로즈업 앵글과 혜수의 팔, 책상의 배치 및 '민숙'이 적힌 휴대폰 화면을 지시사항에 맞게 사실적으로 잘 구현했습니다." }, { "label": "A", "score": 0, "verdict_ko": "인물의 머리가 없는 상태로 경찰 제복만 렌더링된 물리적으로 불가능한 해부학적 오류(하드 위반)가 발생하여 실격입니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지시된 클로즈업 앵글을 유지하며 김 형사를 화면 밖으로 처리하고 휴대폰 화면과 손을 정확히 연출했습니다." }, { "label": "B", "score": 0, "verdict_ko": "인물의 머리가 없는 기형적인 신체가 묘사되는 치명적인 구조적 오류(Hard Violation)가 발생했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 클로즈업 앵글을 유지하며 김 형사를 화면 밖으로 처리하고 휴대폰 화면과 손을 정확히 연출했습니다." }, { "label": "A", "score": 0, "verdict_ko": "인물의 머리가 없는 기형적인 신체가 묘사되는 치명적인 구조적 오류(Hard Violation)가 발생했습니다." } ] }, "combined": { "totals": { "A": 0, "B": 17 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 0, "B": 17 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "요구된 인서트 클로즈업 앵글과 혜수의 팔, 책상의 배치 및 '민숙'이 적힌 휴대폰 화면을 지시사항에 맞게 사실적으로 잘 구현했습니다." }, { "label": "A", "score": 0, "verdict_ko": "인물의 머리가 없는 상태로 경찰 제복만 렌더링된 물리적으로 불가능한 해부학적 오류(하드 위반)가 발생하여 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L13B04.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "화면에 보이는 혜수의 상의가 참고 이미지의 경찰 제복과 달리 견장, 가슴 마크, 'POLICE' 문구가 없는 평범한 남색 점퍼로 묘사되었습니다.", "fix_en": "Modify the visible dark blue jacket to match the police uniform from the reference image by adding the epaulets, chest badge, and 'POLICE' text.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Modify the visible dark blue jacket to match the police uniform from the reference image by adding the epaulets, chest badge, and 'POLICE' text.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 클로즈업 구도와 로케이션을 정확히 구현했으며, 프레임에 노출된 혜수의 경찰복 디테일이 레퍼런스와 완벽하게 일치합니다." }, { "label": "A", "score": 5, "verdict_ko": "요구된 인서트 숏 구도와 배경은 잘 따랐으나, 보이는 팔 부분의 재킷이 레퍼런스의 경찰 유니폼 디테일(패치, 로고 등)을 누락했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 클로즈업 구도와 로케이션을 정확히 구현했으며, 프레임에 노출된 혜수의 경찰복 디테일이 레퍼런스와 완벽하게 일치합니다." }, { "label": "A", "score": 5, "verdict_ko": "요구된 인서트 숏 구도와 배경은 잘 따랐으나, 보이는 팔 부분의 재킷이 레퍼런스의 경찰 유니폼 디테일(패치, 로고 등)을 누락했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 앵글과 샷 크기를 잘 구현했으며, 혜수의 경찰 근무복 디테일(패치와 로고)이 레퍼런스와 정확히 일치하여 더 높은 완성도를 보입니다." }, { "label": "B", "score": 6, "verdict_ko": "구도와 샷 크기는 프롬프트에 부합하나, 인물의 의상에 경찰 마크 및 로고가 누락되어 캐릭터 레퍼런스(의상) 구현에서 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지정된 앵글과 샷 크기를 잘 구현했으며, 혜수의 경찰 근무복 디테일(패치와 로고)이 레퍼런스와 정확히 일치하여 더 높은 완성도를 보입니다." }, { "label": "A", "score": 6, "verdict_ko": "구도와 샷 크기는 프롬프트에 부합하나, 인물의 의상에 경찰 마크 및 로고가 누락되어 캐릭터 레퍼런스(의상) 구현에서 감점되었습니다." } ] }, "combined": { "totals": { "A": 11, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S16sh3__bgfirst_bg.png", "bg_asset_id": "3f8030e6-ea5e-44d9-8d64-1f123ebfdfba", "bg_record_key": "S16sh3::bgfirst_bg", "chain_winner": false, "authority": "plate" }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L13B01", "B": "L13B02", "C": "L13B03", "D": "L13B04" }, "assigned": "L13B04", "choice": "Candidate A", "confident": true, "reason_ko": "휴대폰을 '김 형사 눈앞으로 들이민... 클로즈업' 샷이므로, 책상 앞에 앉은 형사의 1인칭 시점(POV)이나 인물에 근접한 배경을 제공하는 Candidate A가 적합합니다.", "kept": "L13B04" } }, "S17sh2::variants": { "author_fp": "a3cae595648aefff", "author": { "variants": [ { "approach_ko": "얕은 심도로 고양이의 노란 눈과 옆얼굴에 집중하고, 불 켜진 창문은 부드러운 배경 빛으로 남기는 친밀한 관찰 시점.", "prompt_en": "Use an intimate, shallow-focus close-up with the parapet forming a firm lateral base beneath the cat. Render the profile and bright yellow eyes with crisp tactile detail, the eyes carrying a restrained natural glint, while the smaller illuminated window in the upper-right falls softly out of focus but remains clearly recognizable as the endpoint of its gaze. Let subdued dawn ambience skim the black fur and parapet edge, with gentle warm-cool separation from the window light." }, { "approach_ko": "더 깊은 초점과 압축된 공간감으로 고양이, 난간, 상단 우측 창문의 시선 축을 또렷하게 연결하는 긴장된 구도.", "prompt_en": "Photograph the close-up with a more compressed, layered sense of depth and enough focus to keep the parapet’s lateral run and the smaller upper-right window legible behind the cat, while the yellow eyes remain unequivocally the sharpest point. Emphasize the rigid diagonal of the cat’s turned profile toward the window and the narrowness of its perch, creating a taut visual path across the frame. Keep the dawn light low and cool, allowing the illuminated window to register as a contained warm rectangle without overpowering the cat." } ] }, "reused": false }, "S17sh2": { "input_fingerprint": "7cc772e16d3abaa1", "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.\n\nSHOT TEXT (authoritative, Korean): 옥상 난간 위, 불 켜진 옥탑방 창문을 향해 노란 눈을 번뜩이는 검은 고양이의 클로즈업.\n\nLOCATION (lock): On the rooftop parapet outside the rooftop home at dawn, facing its illuminated window. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the black cat’s eye height beside the rooftop boundary, the static close camera holds the cat in profile with its bright yellow eyes as the sharpest point. The cat occupies the lower-center and middle-left while the lit rooftop-room window remains smaller in the background along the direction of its turned head, preserving believable scale before the lateral track begins.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: lit rooftop-room window in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥상 담 (Unobstructed) — The top edge runs laterally beneath the cat, with one side facing the camera; used as Provides the narrow perch beneath the cat and establishes the direction it will cross next; 불 켜진 옥탑방 창문 (Light on) — The exterior-facing side is visible behind and to the right of the cat; used as Creates the background endpoint of the cat’s attention.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dawn ambience remains subdued while the lit window and the cat’s explicitly yellow eyes provide restrained points of brightness within the low-key frame.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains lit through dawn while Min-sook is still absent and the previously erased interior remains undisturbed.\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.\n\nSHOT TEXT (authoritative, Korean): 옥상 난간 위, 불 켜진 옥탑방 창문을 향해 노란 눈을 번뜩이는 검은 고양이의 클로즈업.\n\nLOCATION (lock): On the rooftop parapet outside the rooftop home at dawn, facing its illuminated window. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the black cat’s eye height beside the rooftop boundary, the static close camera holds the cat in profile with its bright yellow eyes as the sharpest point. The cat occupies the lower-center and middle-left while the lit rooftop-room window remains smaller in the background along the direction of its turned head, preserving believable scale before the lateral track begins.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: lit rooftop-room window in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥상 담 (Unobstructed) — The top edge runs laterally beneath the cat, with one side facing the camera; used as Provides the narrow perch beneath the cat and establishes the direction it will cross next; 불 켜진 옥탑방 창문 (Light on) — The exterior-facing side is visible behind and to the right of the cat; used as Creates the background endpoint of the cat’s attention.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dawn ambience remains subdued while the lit window and the cat’s explicitly yellow eyes provide restrained points of brightness within the low-key frame.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains lit through dawn while Min-sook is still absent and the previously erased interior remains undisturbed.\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, shallow-focus close-up with the parapet forming a firm lateral base beneath the cat. Render the profile and bright yellow eyes with crisp tactile detail, the eyes carrying a restrained natural glint, while the smaller illuminated window in the upper-right falls softly out of focus but remains clearly recognizable as the endpoint of its gaze. Let subdued dawn ambience skim the black fur and parapet edge, with gentle warm-cool separation from the window light.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dawn.\n\nSHOT TEXT (authoritative, Korean): 옥상 난간 위, 불 켜진 옥탑방 창문을 향해 노란 눈을 번뜩이는 검은 고양이의 클로즈업.\n\nLOCATION (lock): On the rooftop parapet outside the rooftop home at dawn, facing its illuminated window. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the black cat’s eye height beside the rooftop boundary, the static close camera holds the cat in profile with its bright yellow eyes as the sharpest point. The cat occupies the lower-center and middle-left while the lit rooftop-room window remains smaller in the background along the direction of its turned head, preserving believable scale before the lateral track begins.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: lit rooftop-room window in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 옥상 담 (Unobstructed) — The top edge runs laterally beneath the cat, with one side facing the camera; used as Provides the narrow perch beneath the cat and establishes the direction it will cross next; 불 켜진 옥탑방 창문 (Light on) — The exterior-facing side is visible behind and to the right of the cat; used as Creates the background endpoint of the cat’s attention.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dawn ambience remains subdued while the lit window and the cat’s explicitly yellow eyes provide restrained points of brightness within the low-key frame.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains lit through dawn while Min-sook is still absent and the previously erased interior remains undisturbed.\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 close-up with a more compressed, layered sense of depth and enough focus to keep the parapet’s lateral run and the smaller upper-right window legible behind the cat, while the yellow eyes remain unequivocally the sharpest point. Emphasize the rigid diagonal of the cat’s turned profile toward the window and the narrowness of its perch, creating a taut visual path across the frame. Keep the dawn light low and cool, allowing the illuminated window to register as a contained warm rectangle without overpowering the cat.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": "클로즈업 샷 크기와 프레임 내 피사체의 배치, 노란 눈을 한 고양이의 측면 프로필을 지시문대로 정확히 연출했습니다." }, { "label": "B", "score": 4, "verdict_ko": "요구된 클로즈업 구도를 벗어나 피사체가 작게 잡혔으며, 난간과 옥상 바닥의 구조가 물리적으로 왜곡되는 오류가 발생했습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L03B01.png" }, { "label": "STRUCTURE LOOK — the confirmed photograph of the fixed structure at this location: wherever the structure appears in the frame, its shape, proportions, materials, colors and openings are LOCKED to this photo. Never copy its camera framing, time of day or lighting — the shot text and the LOCATION PHOTOGRAPH are the authorities for those.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/background_chain/seed_bg_rooftop_residence_sel.png" } ], "critique": { "issues": [ { "issue_ko": "고양이가 배경의 불 켜진 창문(오른쪽)을 향하지 않고 반대 방향(왼쪽)을 바라보고 있습니다.", "fix_en": "Turn the cat's head to the right so it is looking backward toward the illuminated window in the background.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Turn the cat’s head and gaze to screen-right toward the illuminated window while preserving its profile and body position.", "issue_ko": "고양이가 화면 왼쪽을 바라보아, 머리 방향의 연장선에 있어야 할 우측 상단의 불 켜진 창문과 시선 축이 반대로 배치되어 있다.", "unfixable": false }, { "fix_en": "Shift the cat slightly left and lower so its body occupies the lower-center and middle-left, keeping the window unobstructed in the upper-right.", "issue_ko": "고양이의 주된 화면 점유가 중간 오른쪽에 치우쳐 있어, 하단 중앙과 중간 왼쪽을 차지해야 한다는 배치와 다르다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Turn the cat's head to the right so it is looking backward toward the illuminated window in the background.\n- Turn the cat’s head and gaze to screen-right toward the illuminated window while preserving its profile and body position.\n- Shift the cat slightly left and lower so its body occupies the lower-center and middle-left, keeping the window unobstructed in the upper-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.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "고양이의 빛나는 노란 눈과 옆모습, 배경의 불 켜진 창문과 난간 등 카메라 구도와 피사체의 조건을 정확히 충족합니다." }, { "label": "B", "score": 3, "verdict_ko": "프롬프트에서 가장 중요하게 요구된 고양이의 노란 눈이 화면에 전혀 나타나지 않았습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "고양이의 빛나는 노란 눈과 옆모습, 배경의 불 켜진 창문과 난간 등 카메라 구도와 피사체의 조건을 정확히 충족합니다." }, { "label": "B", "score": 3, "verdict_ko": "프롬프트에서 가장 중요하게 요구된 고양이의 노란 눈이 화면에 전혀 나타나지 않았습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "프롬프트가 요구한 고양이의 측면 모습과 노란 눈을 정확하게 묘사하여 카메라 프레이밍과 핵심 디테일을 잘 충족했습니다." }, { "label": "A", "score": 3, "verdict_ko": "프롬프트의 핵심인 '노란 눈'이 전혀 보이지 않아 지시사항을 크게 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "프롬프트가 요구한 고양이의 측면 모습과 노란 눈을 정확하게 묘사하여 카메라 프레이밍과 핵심 디테일을 잘 충족했습니다." }, { "label": "B", "score": 3, "verdict_ko": "프롬프트의 핵심인 '노란 눈'이 전혀 보이지 않아 지시사항을 크게 위반했습니다." } ] }, "combined": { "totals": { "A": 15, "B": 6 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트+seed만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_bypass:bg_only", "plate_select": { "candidates": { "A": "L03B01", "B": "L03B02" }, "assigned": "L03B01", "choice": "A", "confident": true, "reason_ko": "지문에서 '옥상 난간 위... 검은 고양이'를 명시하고 있으며, 후보 A에는 난간 위에 앉아 있는 검은 고양이가 정확히 묘사되어 있습니다.", "chosen": "L03B01" } }, "S18sh1::bgfirst_bg": { "input_fingerprint": "0db3441a02cb5064", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 아침 햇살이 비치는 침대 모서리에 이마를 댄 채 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): At the edge of the bed in the main bedroom inside the rooftop home, with early morning daylight entering the room.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From mattress height at the prescribed lateral bedside position, hold a static wide profile of 수리영 folded beside the bed, her head and shoulders low against the right half of the frame while the bed corner anchors the left foreground. Keep the composition still and weighted downward so her eventual awakening can register as a sudden break in the frame rather than a camera reaction.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (수리영 has her forehead resting against its corner) — The mattress edge and corner face the camera in profile beside 수리영; used as The bed corner provides the physical anchor for her exhausted posture and divides the foreground from the room behind.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Morning sunlight reaches the bed in restrained muted tones, with low-key contrast preserving the exhausted stillness.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 아침 햇살이 비치는 침대 모서리에 이마를 댄 채 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): At the edge of the bed in the main bedroom inside the rooftop home, with early morning daylight entering the room.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From mattress height at the prescribed lateral bedside position, hold a static wide profile of 수리영 folded beside the bed, her head and shoulders low against the right half of the frame while the bed corner anchors the left foreground. Keep the composition still and weighted downward so her eventual awakening can register as a sudden break in the frame rather than a camera reaction.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (수리영 has her forehead resting against its corner) — The mattress edge and corner face the camera in profile beside 수리영; used as The bed corner provides the physical anchor for her exhausted posture and divides the foreground from the room behind.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Morning sunlight reaches the bed in restrained muted tones, with low-key contrast preserving the exhausted stillness.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S18sh1__bgfirst_bg.png", "asset_id": "5211b8eb-8515-4bd2-8e39-f9852986faf9", "input_asset_ids": [ "cf5bd3a0-94fa-4d9c-aef0-83403634051b", "afda0f17-d7eb-4277-a1d1-909769467cd4" ] }, "S18sh1": { "input_fingerprint": "2689835352ce5450", "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.\n\nSHOT TEXT (authoritative, Korean): 아침 햇살이 비치는 침대 모서리에 이마를 댄 채 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): At the edge of the bed in the main bedroom inside the rooftop home, with early morning daylight entering the room. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From mattress height at the prescribed lateral bedside position, hold a static wide profile of 수리영 folded beside the bed, her head and shoulders low against the right half of the frame while the bed corner anchors the left foreground. Keep the composition still and weighted downward so her eventual awakening can register as a sudden break in the frame rather than a camera reaction.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (수리영 has her forehead resting against its corner) — The mattress edge and corner face the camera in profile beside 수리영; used as The bed corner provides the physical anchor for her exhausted posture and divides the foreground from the room behind.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Morning sunlight reaches the bed in restrained muted tones, with low-key contrast preserving the exhausted stillness.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook remains missing, and the rooftop room is still free of the corpse and blood that vanished during 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 아침 햇살이 비치는 침대 모서리에 이마를 댄 채 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): At the edge of the bed in the main bedroom inside the rooftop home, with early morning daylight entering the room. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From mattress height at the prescribed lateral bedside position, hold a static wide profile of 수리영 folded beside the bed, her head and shoulders low against the right half of the frame while the bed corner anchors the left foreground. Keep the composition still and weighted downward so her eventual awakening can register as a sudden break in the frame rather than a camera reaction.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (수리영 has her forehead resting against its corner) — The mattress edge and corner face the camera in profile beside 수리영; used as The bed corner provides the physical anchor for her exhausted posture and divides the foreground from the room behind.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Morning sunlight reaches the bed in restrained muted tones, with low-key contrast preserving the exhausted stillness.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook remains missing, and the rooftop room is still free of the corpse and blood that vanished during 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 아침 햇살이 비치는 침대 모서리에 이마를 댄 채 웅크리고 앉아 있는 수리영의 측면.\n\nLOCATION (lock): At the edge of the bed in the main bedroom inside the rooftop home, with early morning daylight entering the room. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From mattress height at the prescribed lateral bedside position, hold a static wide profile of 수리영 folded beside the bed, her head and shoulders low against the right half of the frame while the bed corner anchors the left foreground. Keep the composition still and weighted downward so her eventual awakening can register as a sudden break in the frame rather than a camera reaction.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 침대 (수리영 has her forehead resting against its corner) — The mattress edge and corner face the camera in profile beside 수리영; used as The bed corner provides the physical anchor for her exhausted posture and divides the foreground from the room behind.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Morning sunlight reaches the bed in restrained muted tones, with low-key contrast preserving the exhausted stillness.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook remains missing, and the rooftop room is still free of the corpse and blood that vanished during 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S18sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S18sh1.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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B05.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷 프레이밍과 우측에 웅크린 피사체, 좌측 전경의 침대 모서리 구도를 훌륭하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "피사체가 화면에 너무 크게 잡혀 와이드 샷 프레이밍 요구사항에 다소 어긋나며, 좌측 전경의 구도적 닻 역할이 부족합니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷 프레이밍과 우측에 웅크린 피사체, 좌측 전경의 침대 모서리 구도를 훌륭하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "피사체가 화면에 너무 크게 잡혀 와이드 샷 프레이밍 요구사항에 다소 어긋나며, 좌측 전경의 구도적 닻 역할이 부족합니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 와이드 샷 프레이밍과 침대 모서리에 이마를 댄 웅크린 자세를 완벽하게 구현했으며, 레퍼런스의 후드티 의상과 로케이션 배경도 정확하게 일치합니다." }, { "label": "A", "score": 4, "verdict_ko": "요구된 와이드 샷보다 카메라가 너무 가깝게 잡혀(미디엄 샷) 프레이밍 지시를 어겼으며, 레퍼런스의 후드티 대신 일반 맨투맨을 입고 있어 의상 정확도가 떨어집니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷 프레이밍과 침대 모서리에 이마를 댄 웅크린 자세를 완벽하게 구현했으며, 레퍼런스의 후드티 의상과 로케이션 배경도 정확하게 일치합니다." }, { "label": "B", "score": 4, "verdict_ko": "요구된 와이드 샷보다 카메라가 너무 가깝게 잡혀(미디엄 샷) 프레이밍 지시를 어겼으며, 레퍼런스의 후드티 대신 일반 맨투맨을 입고 있어 의상 정확도가 떨어집니다." } ] }, "combined": { "totals": { "A": 14, "B": 9 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 14, "B": 9 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷 프레이밍과 우측에 웅크린 피사체, 좌측 전경의 침대 모서리 구도를 훌륭하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "피사체가 화면에 너무 크게 잡혀 와이드 샷 프레이밍 요구사항에 다소 어긋나며, 좌측 전경의 구도적 닻 역할이 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B05.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "캐릭터가 발바닥을 바닥에 대고 쪼그려 앉아 있으나, 레이아웃 스케치에서는 무릎을 바닥에 대고 앉아 있습니다.", "fix_en": "Adjust the character's lower body so her knees are resting flat on the floor and her hips sit back on her legs, matching the kneeling pose in the layout sketch.", "unfixable": false }, { "issue_ko": "원본 배경 이미지에 있던 침대 옆 작은 협탁과 그 위의 시계가 캐릭터 등 뒤에서 삭제되었습니다.", "fix_en": "Restore the small wooden nightstand, clock, and cup from the original background reference exactly where they belong, visible behind the character.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Preserve the existing geometry; the shifted camera axis and altered background framing cannot be corrected in place.", "issue_ko": "카메라 시점과 화면 축이 지정된 빈 배경과 일치하지 않아 천장등, 옷장, 출입문, 창문의 화면 위치와 침대의 전경 비중이 달라졌다.", "unfixable": true }, { "fix_en": "Repose her from a compact foot-supported squat into the sketched low, cross-legged seated curl, extending her folded legs toward screen right while keeping her forehead fixed against the bed corner.", "issue_ko": "수리영이 스케치처럼 바닥에 낮게 다리를 접어 앉은 자세가 아니라 발로 지탱한 쪼그려 앉기 자세여서 몸이 지나치게 작고 좁게 뭉쳐 있다.", "unfixable": false }, { "fix_en": "Move her forearms and hands from her chest to the sketched position beside her lowered face and the bed edge, preserving her head contact and all other staging.", "issue_ko": "양손과 팔이 스케치의 얼굴 옆·침대 모서리 부근이 아니라 가슴 앞에 모여 있어 지정된 웅크린 실루엣과 팔 배치가 다르다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Adjust the character's lower body so her knees are resting flat on the floor and her hips sit back on her legs, matching the kneeling pose in the layout sketch.\n- Restore the small wooden nightstand, clock, and cup from the original background reference exactly where they belong, visible behind the character.\n- Repose her from a compact foot-supported squat into the sketched low, cross-legged seated curl, extending her folded legs toward screen right while keeping her forehead fixed against the bed corner.\n- Move her forearms and hands from her chest to the sketched position beside her lowered face and the bed edge, preserving her head contact and all other staging.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "원본 배경의 모든 요소(협탁 등)를 온전히 유지했으며, 스케치에 제시된 무릎을 꿇은 자세와 프레이밍을 가장 충실하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "원본 배경에 있던 협탁과 소품들이 완전히 삭제되는 치명적인 오류가 발생했으며, 스케치와 달리 발바닥을 대고 쭈그려 앉은 자세로 변형되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "원본 배경의 모든 요소(협탁 등)를 온전히 유지했으며, 스케치에 제시된 무릎을 꿇은 자세와 프레이밍을 가장 충실하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "원본 배경에 있던 협탁과 소품들이 완전히 삭제되는 치명적인 오류가 발생했으며, 스케치와 달리 발바닥을 대고 쭈그려 앉은 자세로 변형되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "스케치의 무릎 꿇은 자세 대신 쪼그려 앉았으나, 팔을 가슴에 모으고 침대에 직접 이마를 댄 상체 포즈가 텍스트 지시와 스케치에 훨씬 더 부합합니다." }, { "label": "A", "score": 5, "verdict_ko": "하체는 스케치와 일치하지만, 팔을 앞으로 뻗어 손 위에 이마를 얹음으로써 '침대에 이마를 댄' 텍스트 지시와 스케치의 상체 포즈를 위반했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "스케치의 무릎 꿇은 자세 대신 쪼그려 앉았으나, 팔을 가슴에 모으고 침대에 직접 이마를 댄 상체 포즈가 텍스트 지시와 스케치에 훨씬 더 부합합니다." }, { "label": "B", "score": 5, "verdict_ko": "하체는 스케치와 일치하지만, 팔을 앞으로 뻗어 손 위에 이마를 얹음으로써 '침대에 이마를 댄' 텍스트 지시와 스케치의 상체 포즈를 위반했습니다." } ] }, "combined": { "totals": { "A": 10, "B": 12 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S18sh1__bgfirst_bg.png", "bg_asset_id": "5211b8eb-8515-4bd2-8e39-f9852986faf9", "bg_record_key": "S18sh1::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B05", "choice": "Candidate E", "confident": true, "reason_ko": "\"아침 햇살이 비치는 침대 모서리에\"", "kept": "L04B05" } }, "S18sh10::variants": { "author_fp": "4636a5df30cb8340", "author": { "variants": [ { "approach_ko": "정면 대칭과 깊은 초점으로 붉은 원을 냉정하고 증거물처럼 고립시키는 접근.", "prompt_en": "Photograph the wall with a restrained, nearly orthographic lens feel, strict frontal symmetry, and deep focus across the circle, tile faces, grout lines, and surrounding surface. Use soft, subdued morning bathroom light with minimal shadow, letting the physically present red stroke stand in stark, ominous isolation against the muted tile." }, { "approach_ko": "비스듬히 스치는 빛과 얕은 질감 강조로 타일 위에 떠오른 붉은 원의 물성을 불길하게 드러내는 접근.", "prompt_en": "Keep the square-on insert composition but give it a more intimate, tactile lens rendering, with the complete red ring crisply resolved while the surrounding tiles fall away gently toward the frame edges. Let subdued morning illumination rake softly across the wall, revealing grout relief, tile sheen, and the red stroke sitting physically on the surface, with controlled peripheral dimness concentrating unease at the center." } ] }, "reused": false }, "S18sh10": { "input_fingerprint": "b175131beee01cd9", "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.\n\nSHOT TEXT (authoritative, Korean): 거울 뒤 타일 벽면, 불길한 붉은 원 형상이 뚜렷하게 떠오른 모습.\n\nLOCATION (lock): The tiled wall behind the mirror inside the rooftop home's bathroom, lit for morning washing. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the terminal point of the inherited eye-level dolly, press into an insert of the bathroom tile wall and crop away nearly all of 수리영 and the mirror. Hold the fully formed red circle near the middle of the frame with enough surrounding tile to establish that the mark occupies the wall rather than the lens.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: red circle on bathroom tile wall in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 욕실 타일 벽면 (A distinct red circle has appeared on it) — The camera faces the marked side squarely, with the complete red circle visible on that surface; used as The wall becomes the isolated focal plane at the end of the dolly.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued bathroom illumination keeps the palette muted while the manifested red circle supplies the sole deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The home remains unnaturally clean and corpse-free, and Su-ri-young has taken out her doll-adorned backpack. The red circle on the bathroom wall is only a momentary apparition and leaves no lasting mark.\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.\n\nSHOT TEXT (authoritative, Korean): 거울 뒤 타일 벽면, 불길한 붉은 원 형상이 뚜렷하게 떠오른 모습.\n\nLOCATION (lock): The tiled wall behind the mirror inside the rooftop home's bathroom, lit for morning washing. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the terminal point of the inherited eye-level dolly, press into an insert of the bathroom tile wall and crop away nearly all of 수리영 and the mirror. Hold the fully formed red circle near the middle of the frame with enough surrounding tile to establish that the mark occupies the wall rather than the lens.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: red circle on bathroom tile wall in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 욕실 타일 벽면 (A distinct red circle has appeared on it) — The camera faces the marked side squarely, with the complete red circle visible on that surface; used as The wall becomes the isolated focal plane at the end of the dolly.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued bathroom illumination keeps the palette muted while the manifested red circle supplies the sole deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The home remains unnaturally clean and corpse-free, and Su-ri-young has taken out her doll-adorned backpack. The red circle on the bathroom wall is only a momentary apparition and leaves no lasting mark.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the wall with a restrained, nearly orthographic lens feel, strict frontal symmetry, and deep focus across the circle, tile faces, grout lines, and surrounding surface. Use soft, subdued morning bathroom light with minimal shadow, letting the physically present red stroke stand in stark, ominous isolation against the muted tile.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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.\n\nSHOT TEXT (authoritative, Korean): 거울 뒤 타일 벽면, 불길한 붉은 원 형상이 뚜렷하게 떠오른 모습.\n\nLOCATION (lock): The tiled wall behind the mirror inside the rooftop home's bathroom, lit for morning washing. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At the terminal point of the inherited eye-level dolly, press into an insert of the bathroom tile wall and crop away nearly all of 수리영 and the mirror. Hold the fully formed red circle near the middle of the frame with enough surrounding tile to establish that the mark occupies the wall rather than the lens.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: red circle on bathroom tile wall in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 욕실 타일 벽면 (A distinct red circle has appeared on it) — The camera faces the marked side squarely, with the complete red circle visible on that surface; used as The wall becomes the isolated focal plane at the end of the dolly.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued bathroom illumination keeps the palette muted while the manifested red circle supplies the sole deep-red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The home remains unnaturally clean and corpse-free, and Su-ri-young has taken out her doll-adorned backpack. The red circle on the bathroom wall is only a momentary apparition and leaves no lasting mark.\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):\nKeep the square-on insert composition but give it a more intimate, tactile lens rendering, with the complete red ring crisply resolved while the surrounding tiles fall away gently toward the frame edges. Let subdued morning illumination rake softly across the wall, revealing grout relief, tile sheen, and the red stroke sitting physically on the surface, with controlled peripheral dimness concentrating unease at the center.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 9, "A": 7 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "지시된 타일 벽면의 인서트 클로즈업 구도를 완벽히 구현했으며, 차분하고 어두운 조명 톤이 '불길한' 분위기와 짙은 붉은색 액센트를 잘 살려냅니다." }, { "label": "A", "score": 7, "verdict_ko": "타일 벽면과 붉은 원의 클로즈업 구도는 정확히 따랐으나, 조명이 다소 밝아 프롬프트가 요구한 '차분하고 음울한' 무드가 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B04.png" } ], "critique": { "issues": [ { "issue_ko": "붉은 원이 타일 줄눈 부분에서 끊어져 네 조각으로 나뉘어 있습니다; 끊어짐 없이 연속된 하나의 둥근 선으로 이어져야 합니다.", "fix_en": "Connect the four disconnected red arcs across the tile grout lines to form a single, continuous circular brushstroke.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Connect the four disconnected red arcs across the tile grout lines to form a single, continuous circular brushstroke.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "프롬프트가 요구한 '붉은 원' 형상과 타일 벽면의 인서트 샷을 정확히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "원 내부에 십자선이 추가되어 프롬프트가 지시한 단순한 '원 형상'의 범위를 벗어났습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "프롬프트가 요구한 '붉은 원' 형상과 타일 벽면의 인서트 샷을 정확히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "원 내부에 십자선이 추가되어 프롬프트가 지시한 단순한 '원 형상'의 범위를 벗어났습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "타일 벽면 중앙에 지시된 붉은 원 형상만을 선명하게 담아낸 인서트 샷 구도를 정확히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "요구된 붉은 원 외에 내부에 지시문에 없는 십자 형태의 선이 임의로 추가되어 감점되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "타일 벽면 중앙에 지시된 붉은 원 형상만을 선명하게 담아낸 인서트 샷 구도를 정확히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "요구된 붉은 원 외에 내부에 지시문에 없는 십자 형태의 선이 임의로 추가되어 감점되었습니다." } ] }, "combined": { "totals": { "A": 14, "B": 8 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B04", "choice": "Candidate D", "confident": true, "reason_ko": "거울 뒤 타일 벽면이라는 묘사가 등장하므로 거울과 타일이 있는 화장실 공간을 보여주는 후보가 적합합니다.", "kept": "L04B04" } }, "S18sh13::variants": { "author_fp": "7ee5a7e757e9e53e", "author": { "variants": [ { "approach_ko": "정면 대칭과 절제된 깊은 초점으로 평범한 가족사진 속 변형을 냉정하게 응시한다.", "prompt_en": "Use a rigid, dead-on composition with restrained normal-lens perspective, holding the smiling face precisely at the visual center. Keep the nearby family-photo context legible in moderately deep focus while the narrow table edge remains a quiet spatial anchor. Soft morning ambience falls evenly across the photographic surface, making the dark downward deformation feel stark and inexplicable within an otherwise ordinary image." }, { "approach_ko": "얕은 초점과 표면 질감 중심의 촬영으로 먹물 같은 하강 변형을 촉각적이고 불안하게 부각한다.", "prompt_en": "Photograph the image surface with a compressed, intimate lens feel and very shallow depth of field, placing critical focus on the boundary where the smile deforms into descending ink-like trails. Let the surrounding portion of the photograph and the table edge recede gently out of focus without losing their contextual shapes. Use the existing morning daylight at a subtle grazing angle to reveal paper sheen, frame reflections, and the viscous-looking darkness of the transformation." } ] }, "reused": false }, "S18sh13": { "input_fingerprint": "4e977760558c6763", "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.\n\nSHOT TEXT (authoritative, Korean): 사진 속 활짝 웃는 강민숙의 얼굴 부분이 기괴하게 일그러지며 먹물처럼 흘러내리는 찰나.\n\nLOCATION (lock): At the dining table in the rooftop home's living area, illuminated by morning daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the chest-height dolly directly before the framed photograph, tightening to a close view of 강민숙's smiling face while retaining a narrow portion of the picture around her as context. Center the facial deformation as it pulls downward within the photographic image, without enlarging the frame or table beyond realistic scale.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 강민숙과 수리영의 액자 사진 (강민숙's face appears distorted and flowing downward) — Its image-bearing front faces the camera, tightened around 강민숙's smiling face and the surrounding family-photo context; used as The photograph is the focal surface through which the impossible facial change is perceived; 식탁 (The framed photograph rests on it) — Only the upper edge beneath the photograph is visible to camera; used as A narrow lower-frame reference establishes the photograph's real placement in the living room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient living-room light remains restrained and natural to the interior, emphasizing the dark downward distortion without introducing a new light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The family photograph remains on the table and returns to normal after the brief distortion; Su-ri-young still has her doll-adorned backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 사진 속 활짝 웃는 강민숙의 얼굴 부분이 기괴하게 일그러지며 먹물처럼 흘러내리는 찰나.\n\nLOCATION (lock): At the dining table in the rooftop home's living area, illuminated by morning daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the chest-height dolly directly before the framed photograph, tightening to a close view of 강민숙's smiling face while retaining a narrow portion of the picture around her as context. Center the facial deformation as it pulls downward within the photographic image, without enlarging the frame or table beyond realistic scale.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 강민숙과 수리영의 액자 사진 (강민숙's face appears distorted and flowing downward) — Its image-bearing front faces the camera, tightened around 강민숙's smiling face and the surrounding family-photo context; used as The photograph is the focal surface through which the impossible facial change is perceived; 식탁 (The framed photograph rests on it) — Only the upper edge beneath the photograph is visible to camera; used as A narrow lower-frame reference establishes the photograph's real placement in the living room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient living-room light remains restrained and natural to the interior, emphasizing the dark downward distortion without introducing a new light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The family photograph remains on the table and returns to normal after the brief distortion; Su-ri-young still has her doll-adorned backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a rigid, dead-on composition with restrained normal-lens perspective, holding the smiling face precisely at the visual center. Keep the nearby family-photo context legible in moderately deep focus while the narrow table edge remains a quiet spatial anchor. Soft morning ambience falls evenly across the photographic surface, making the dark downward deformation feel stark and inexplicable within an otherwise ordinary 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.", "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.\n\nSHOT TEXT (authoritative, Korean): 사진 속 활짝 웃는 강민숙의 얼굴 부분이 기괴하게 일그러지며 먹물처럼 흘러내리는 찰나.\n\nLOCATION (lock): At the dining table in the rooftop home's living area, illuminated by morning daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Complete the chest-height dolly directly before the framed photograph, tightening to a close view of 강민숙's smiling face while retaining a narrow portion of the picture around her as context. Center the facial deformation as it pulls downward within the photographic image, without enlarging the frame or table beyond realistic scale.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 강민숙과 수리영의 액자 사진 (강민숙's face appears distorted and flowing downward) — Its image-bearing front faces the camera, tightened around 강민숙's smiling face and the surrounding family-photo context; used as The photograph is the focal surface through which the impossible facial change is perceived; 식탁 (The framed photograph rests on it) — Only the upper edge beneath the photograph is visible to camera; used as A narrow lower-frame reference establishes the photograph's real placement in the living room.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Ambient living-room light remains restrained and natural to the interior, emphasizing the dark downward distortion without introducing a new light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The family photograph remains on the table and returns to normal after the brief distortion; Su-ri-young still has her doll-adorned backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the image surface with a compressed, intimate lens feel and very shallow depth of field, placing critical focus on the boundary where the smile deforms into descending ink-like trails. Let the surrounding portion of the photograph and the table edge recede gently out of focus without losing their contextual shapes. Use the existing morning daylight at a subtle grazing angle to reveal paper sheen, frame reflections, and the viscous-looking darkness of the transformation.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "요구된 정면 카메라 구도와 먹물처럼 흘러내리는 효과를 잘 구현했으나, 프레이밍이 다소 넓어 식탁 하단이 과하게 노출되었습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍의 크기는 적절하나, 카메라를 사진 정면에 배치하라는 명확한 구도 지시를 무시하고 측면에서 촬영하여 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B04.png" } ], "critique": { "issues": [ { "issue_ko": "흘러내리는 검은 먹물 효과가 사진 속 평면 이미지에 머물지 않고, 실제 3D 액체처럼 액자 틀을 넘어 식탁 위까지 물리적으로 흘러넘치고 있습니다.", "fix_en": "Confine the black melting effect strictly to the 2D surface of the photograph, removing the 3D liquid that spills over the wooden frame and pools on the table.", "unfixable": false }, { "issue_ko": "인물 얼굴에 맞춘 클로즈업 샷이어야 하며 식탁은 사진 아래 좁은 가장자리만 보여야 하지만, 카메라 구도가 너무 넓어 액자 전체와 깊은 식탁 표면이 모두 노출되었습니다.", "fix_en": "The camera framing is too wide, showing the full picture frame and deep table surface instead of a tight close-up on the face; this cannot be corrected without moving the camera.", "unfixable": true } ] }, "composition_critique": { "issues": [ { "fix_en": "Crop tightly into the photograph around Kang Min-sook’s smiling face, place her face at the exact visual center, retain only a narrow band of surrounding family-photo context, and leave only a thin table edge beneath it.", "issue_ko": "구도가 지나치게 넓어 액자 전체와 넓은 식탁 및 거실 배경이 보이며, 강민숙의 얼굴이 화면 정중앙을 차지하는 밀착 클로즈업이 아니다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Confine the black melting effect strictly to the 2D surface of the photograph, removing the 3D liquid that spills over the wooden frame and pools on the table.\n- Crop tightly into the photograph around Kang Min-sook’s smiling face, place her face at the exact visual center, retain only a narrow band of surrounding family-photo context, and leave only a thin table edge beneath it.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 앵글과 스케일로 식탁 위 액자를 정확히 포착했으며, 얼굴이 먹물처럼 기괴하게 흘러내리는 묘사가 프롬프트와 잘 일치합니다." }, { "label": "B", "score": 7, "verdict_ko": "후보 A와 동일한 이미지로, 요구된 클로즈업 샷과 기괴한 변형, 배경의 일치도를 모두 훌륭하게 충족합니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 앵글과 스케일로 식탁 위 액자를 정확히 포착했으며, 얼굴이 먹물처럼 기괴하게 흘러내리는 묘사가 프롬프트와 잘 일치합니다." }, { "label": "B", "score": 7, "verdict_ko": "후보 A와 동일한 이미지로, 요구된 클로즈업 샷과 기괴한 변형, 배경의 일치도를 모두 훌륭하게 충족합니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 클로즈업 구도와 배경 요소를 정확히 구현했으며, 액자가 식탁 위에 자연스럽게 배치되었습니다." }, { "label": "B", "score": 6, "verdict_ko": "대부분의 지시사항을 잘 따랐으나, 액자 하단과 식탁 사이에 불필요한 검은색 구조물이 생성되어 A보다 약간 부자연스럽습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 클로즈업 구도와 배경 요소를 정확히 구현했으며, 액자가 식탁 위에 자연스럽게 배치되었습니다." }, { "label": "A", "score": 6, "verdict_ko": "대부분의 지시사항을 잘 따랐으나, 액자 하단과 식탁 사이에 불필요한 검은색 구조물이 생성되어 A보다 약간 부자연스럽습니다." } ] }, "combined": { "totals": { "A": 13, "B": 14 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B04", "choice": "Candidate A", "confident": true, "reason_ko": "로케이션 설명에 '아침 햇살이 비치는 거실의 식탁(At the dining table in the rooftop home's living area, illuminated by morning daylight)'이라고 명시되어 있으므로, 화장실인 기존 할당 컷이 아닌 주간의 거실 식탁 공간이 정면으로 보이는 이미지가 적합합니다.", "kept": "L04B04" } }, "S19sh1::bgfirst_bg": { "input_fingerprint": "79e8d0d709d50102", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한산한 아침의 마트 입구, 세워진 자전거 곁에서 마트 안쪽 한 지점을 바라보는 수리영의 전신.\n\nLOCATION (lock): Directly outside the large supermarket's entrance on a quiet morning, beside the bicycle parking spot.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level several meters before the market entrance, begin with a full-body view of 수리영 beside the parked bicycle, her back and turning shoulder leading toward the store interior behind her. Leave open depth beyond her so the frame is ready to track backward through the entrance when she advances.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (수리영 is about to move inside) — The entrance faces the camera, with the interior receding behind 수리영; used as The entrance creates the deep background route that her body orientation points into; 자전거 (Parked beside the entrance) — Its side is visible next to 수리영's full figure; used as The bicycle remains beside her as a grounded scale reference before she enters.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Early-morning ambient light is rendered in muted neutrals with restrained contrast appropriate to the quiet market entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 한산한 아침의 마트 입구, 세워진 자전거 곁에서 마트 안쪽 한 지점을 바라보는 수리영의 전신.\n\nLOCATION (lock): Directly outside the large supermarket's entrance on a quiet morning, beside the bicycle parking spot.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level several meters before the market entrance, begin with a full-body view of 수리영 beside the parked bicycle, her back and turning shoulder leading toward the store interior behind her. Leave open depth beyond her so the frame is ready to track backward through the entrance when she advances.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (수리영 is about to move inside) — The entrance faces the camera, with the interior receding behind 수리영; used as The entrance creates the deep background route that her body orientation points into; 자전거 (Parked beside the entrance) — Its side is visible next to 수리영's full figure; used as The bicycle remains beside her as a grounded scale reference before she enters.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Early-morning ambient light is rendered in muted neutrals with restrained contrast appropriate to the quiet market entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S19sh1__bgfirst_bg.png", "asset_id": "a45c8b93-be82-4cbb-b44e-25ffa7c9dd85", "input_asset_ids": [ "0e690936-d9ba-4651-8977-cd180ccb7163", "b83e7112-5660-4f1d-9503-083c0aa89ca1", "40c8635a-28fd-4d3f-b913-166aac4c1a19" ] }, "S19sh1": { "input_fingerprint": "67f4b96b16de1339", "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.\n\nSHOT TEXT (authoritative, Korean): 한산한 아침의 마트 입구, 세워진 자전거 곁에서 마트 안쪽 한 지점을 바라보는 수리영의 전신.\n\nLOCATION (lock): Directly outside the large supermarket's entrance on a quiet morning, beside the bicycle parking spot. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level several meters before the market entrance, begin with a full-body view of 수리영 beside the parked bicycle, her back and turning shoulder leading toward the store interior behind her. Leave open depth beyond her so the frame is ready to track backward through the entrance when she advances.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (수리영 is about to move inside) — The entrance faces the camera, with the interior receding behind 수리영; used as The entrance creates the deep background route that her body orientation points into; 자전거 (Parked beside the entrance) — Its side is visible next to 수리영's full figure; used as The bicycle remains beside her as a grounded scale reference before she enters.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Early-morning ambient light is rendered in muted neutrals with restrained contrast appropriate to the quiet market entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s bicycle remains parked at the mart entrance, and she continues carrying the doll-adorned backpack taken from home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 한산한 아침의 마트 입구, 세워진 자전거 곁에서 마트 안쪽 한 지점을 바라보는 수리영의 전신.\n\nLOCATION (lock): Directly outside the large supermarket's entrance on a quiet morning, beside the bicycle parking spot. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level several meters before the market entrance, begin with a full-body view of 수리영 beside the parked bicycle, her back and turning shoulder leading toward the store interior behind her. Leave open depth beyond her so the frame is ready to track backward through the entrance when she advances.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (수리영 is about to move inside) — The entrance faces the camera, with the interior receding behind 수리영; used as The entrance creates the deep background route that her body orientation points into; 자전거 (Parked beside the entrance) — Its side is visible next to 수리영's full figure; used as The bicycle remains beside her as a grounded scale reference before she enters.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Early-morning ambient light is rendered in muted neutrals with restrained contrast appropriate to the quiet market entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s bicycle remains parked at the mart entrance, and she continues carrying the doll-adorned backpack taken from home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 한산한 아침의 마트 입구, 세워진 자전거 곁에서 마트 안쪽 한 지점을 바라보는 수리영의 전신.\n\nLOCATION (lock): Directly outside the large supermarket's entrance on a quiet morning, beside the bicycle parking spot. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level several meters before the market entrance, begin with a full-body view of 수리영 beside the parked bicycle, her back and turning shoulder leading toward the store interior behind her. Leave open depth beyond her so the frame is ready to track backward through the entrance when she advances.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: 마트 입구 (수리영 is about to move inside) — The entrance faces the camera, with the interior receding behind 수리영; used as The entrance creates the deep background route that her body orientation points into; 자전거 (Parked beside the entrance) — Its side is visible next to 수리영's full figure; used as The bicycle remains beside her as a grounded scale reference before she enters.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Early-morning ambient light is rendered in muted neutrals with restrained contrast appropriate to the quiet market entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s bicycle remains parked at the mart entrance, and she continues carrying the doll-adorned backpack taken from home.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S19sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S19sh1.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L09B01.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/79670d0c-2096-4d40-88cf-30b407411548/images/background_chain/seed_bg_retail_store_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": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "마트 안쪽을 바라보는 뒷모습과 지시된 시선 방향을 정확히 연출했으며, 자전거와 인형이 달린 백팩 등 요구된 디테일을 잘 반영했습니다." }, { "label": "A", "score": 7, "verdict_ko": "캐릭터가 마트 안쪽이 아닌 화면 우측을 바라보고 있어 '마트 안쪽 한 지점을 바라보는'이라는 핵심 행동 지시를 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "마트 안쪽을 바라보는 뒷모습과 지시된 시선 방향을 정확히 연출했으며, 자전거와 인형이 달린 백팩 등 요구된 디테일을 잘 반영했습니다." }, { "label": "A", "score": 7, "verdict_ko": "캐릭터가 마트 안쪽이 아닌 화면 우측을 바라보고 있어 '마트 안쪽 한 지점을 바라보는'이라는 핵심 행동 지시를 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "마트 안을 바라보기 위해 어깨를 돌린 측후면 포즈(turning shoulder), 인물과 배경의 스케일, 인형이 달린 배낭과 자전거의 위치 등 샷 텍스트와 프레이밍 지시사항을 매우 정확하고 자연스럽게 구현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "지정된 위치와 인물, 소품의 조건은 갖추었으나, '어깨를 돌려 마트 안쪽을 향하는' 지시와 달리 카메라를 완전히 등진 평면적인 뒷모습으로 연출되어 프롬프트의 디테일을 놓쳤습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "마트 안을 바라보기 위해 어깨를 돌린 측후면 포즈(turning shoulder), 인물과 배경의 스케일, 인형이 달린 배낭과 자전거의 위치 등 샷 텍스트와 프레이밍 지시사항을 매우 정확하고 자연스럽게 구현했습니다." }, { "label": "B", "score": 7, "verdict_ko": "지정된 위치와 인물, 소품의 조건은 갖추었으나, '어깨를 돌려 마트 안쪽을 향하는' 지시와 달리 카메라를 완전히 등진 평면적인 뒷모습으로 연출되어 프롬프트의 디테일을 놓쳤습니다." } ] }, "combined": { "totals": { "A": 17, "B": 16 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 17, "B": 16 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "마트 안쪽을 바라보는 뒷모습과 지시된 시선 방향을 정확히 연출했으며, 자전거와 인형이 달린 백팩 등 요구된 디테일을 잘 반영했습니다." }, { "label": "A", "score": 7, "verdict_ko": "캐릭터가 마트 안쪽이 아닌 화면 우측을 바라보고 있어 '마트 안쪽 한 지점을 바라보는'이라는 핵심 행동 지시를 위반했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L09B01.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/79670d0c-2096-4d40-88cf-30b407411548/images/background_chain/seed_bg_retail_store_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "자전거가 원본 배경 및 레이아웃 스케치와 달리 거치대 안에 꽂혀 있지 않고 거치대 뒤쪽에 떠 있는 형태로 배치되어 있습니다.", "fix_en": "Reposition the bicycle so its front wheel is resting inside the metal parking rack, exactly matching its placement in the original background reference.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Restore the bicycle and bicycle rack to their exact screen positions and scales from the shot background, shifting them right and reducing the bicycle’s apparent size while preserving the woman and all other geometry.", "issue_ko": "자전거와 자전거 거치대가 고정 배경의 위치보다 화면 왼쪽으로 이동했고, 자전거도 원본보다 전경에서 더 크게 보여 배경 잠금 구도와 어긋난다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Reposition the bicycle so its front wheel is resting inside the metal parking rack, exactly matching its placement in the original background reference.\n- Restore the bicycle and bicycle rack to their exact screen positions and scales from the shot background, shifting them right and reducing the bicycle’s apparent size while preserving the woman and all other geometry.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 원본 배경을 그대로 유지하면서 인물의 위치와 자세, 가방 등의 디테일을 스케치와 프롬프트에 맞게 정확히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "원본 배경 이미지에서 좌측 거치대에 있던 자전거의 위치를 우측으로 임의로 이동시켜, 배경을 정확히 유지하라는 지침을 위반했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 원본 배경을 그대로 유지하면서 인물의 위치와 자세, 가방 등의 디테일을 스케치와 프롬프트에 맞게 정확히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "원본 배경 이미지에서 좌측 거치대에 있던 자전거의 위치를 우측으로 임의로 이동시켜, 배경을 정확히 유지하라는 지침을 위반했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "배경 원본을 그대로 유지해야 한다는 조건을 어기고 자전거의 위치를 오른쪽으로 임의 변경하여 심각한 감점 요소가 되었습니다." }, { "label": "B", "score": 8, "verdict_ko": "지정된 배경과 자전거 위치를 완벽히 유지하면서 레이아웃 스케치에 맞게 인물의 구도와 디테일을 충실히 구현했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 3, "verdict_ko": "배경 원본을 그대로 유지해야 한다는 조건을 어기고 자전거의 위치를 오른쪽으로 임의 변경하여 심각한 감점 요소가 되었습니다." }, { "label": "A", "score": 8, "verdict_ko": "지정된 배경과 자전거 위치를 완벽히 유지하면서 레이아웃 스케치에 맞게 인물의 구도와 디테일을 충실히 구현했습니다." } ] }, "combined": { "totals": { "A": 16, "B": 7 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S19sh1__bgfirst_bg.png", "bg_asset_id": "a45c8b93-be82-4cbb-b44e-25ffa7c9dd85", "bg_record_key": "S19sh1::bgfirst_bg", "chain_winner": true, "authority": "plate", "seed_attached": true }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready", "plate_select": { "candidates": { "A": "L09B01", "B": "L09B02" }, "assigned": "L09B01", "choice": "Candidate A", "confident": true, "reason_ko": "지문에서 '마트 입구, 세워진 자전거 곁'을 명시하고 있으며, 두 후보 모두 해당 동일한 공간을 보여주고 있으므로 명확한 변경 근거가 없어 현재 할당된 이미지를 유지합니다.", "kept": "L09B01" } }, "S19sh3::variants": { "author_fp": "494fe5f7a9fa837c", "author": { "variants": [ { "approach_ko": "깊은 초점과 외벽의 수직 구조로 수리영의 시선과 CCTV 사이의 증거적 연결을 또렷하게 강조한다.", "prompt_en": "Use a restrained, observational medium profile with a natural lens feel and generous depth of field, keeping both Su-ri-young’s raised profile and the elevated CCTV housing clearly resolved. Let the exterior wall’s vertical plane dominate the composition, while its lines and the open negative space make the lower-left-to-upper-right eyeline diagonal unmistakable. Neutral early-morning light falls evenly across her face, clothing, backpack, wall texture, and camera housing, with subdued contrast and precise material detail." }, { "approach_ko": "얕은 심도와 친밀한 측면 관찰로 수리영의 경계 어린 시선을 중심에 두고 CCTV를 압박감 있는 감시 대상으로 만든다.", "prompt_en": "Photograph the prescribed medium profile with a subtly compressed lens feel and selective depth, holding crisp attention on Su-ri-young’s eye, cheek, and upward-tilted posture while the upper-right CCTV remains distinctly recognizable just beyond the primary focus plane. Use the wall as a quiet, minimally textured field so the visual tension travels directly along her gaze to the surveillance camera. Shape the subdued early-morning light softly across her profile, with gentle falloff toward the elevated fixture and controlled, natural contrast." } ] }, "reused": false }, "S19sh3": { "input_fingerprint": "2dca2b29423c600e", "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.\n\nSHOT TEXT (authoritative, Korean): 마트 밖, 건물 외벽에 매달린 CCTV 카메라를 올려다보는 수리영의 측면.\n\nLOCATION (lock): Outside the supermarket, beneath a CCTV camera mounted high on the building's exterior wall. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the prescribed chest-height side axis, hold a medium profile of 수리영 in the lower-left portion while the wall-mounted CCTV camera sits above and across her eyeline in the upper-right. Freeze the tilt at its upper extent so the diagonal between her raised gaze and the camera is physically clear before the next downward move.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, midground, looks toward wall-mounted CCTV camera; wall-mounted CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 외벽 CCTV 카메라 (Mounted on the market exterior wall) — Its camera-facing housing and lens direction are visible above 수리영's sightline; used as The surveillance camera forms the upper endpoint of 수리영's eyeline and the evidentiary focus of the shot; 마트 외벽 (Supports the CCTV camera); used as The wall provides the vertical plane connecting 수리영 and the elevated camera.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Early-morning exterior light remains neutral and subdued, with controlled contrast linking the figure to the surveillance camera.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same morning light, exterior wall materials, entrance facade, and quiet storefront atmosphere from the reference. Exclude the parked bicycle and the interior aisles if they fall outside this upward view, and show only the woman and the wall-mounted security camera.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s bicycle remains parked outside the mart while she checks the exterior CCTV position, 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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.\n\nSHOT TEXT (authoritative, Korean): 마트 밖, 건물 외벽에 매달린 CCTV 카메라를 올려다보는 수리영의 측면.\n\nLOCATION (lock): Outside the supermarket, beneath a CCTV camera mounted high on the building's exterior wall. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the prescribed chest-height side axis, hold a medium profile of 수리영 in the lower-left portion while the wall-mounted CCTV camera sits above and across her eyeline in the upper-right. Freeze the tilt at its upper extent so the diagonal between her raised gaze and the camera is physically clear before the next downward move.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, midground, looks toward wall-mounted CCTV camera; wall-mounted CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 외벽 CCTV 카메라 (Mounted on the market exterior wall) — Its camera-facing housing and lens direction are visible above 수리영's sightline; used as The surveillance camera forms the upper endpoint of 수리영's eyeline and the evidentiary focus of the shot; 마트 외벽 (Supports the CCTV camera); used as The wall provides the vertical plane connecting 수리영 and the elevated camera.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Early-morning exterior light remains neutral and subdued, with controlled contrast linking the figure to the surveillance camera.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same morning light, exterior wall materials, entrance facade, and quiet storefront atmosphere from the reference. Exclude the parked bicycle and the interior aisles if they fall outside this upward view, and show only the woman and the wall-mounted security camera.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s bicycle remains parked outside the mart while she checks the exterior CCTV position, 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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, observational medium profile with a natural lens feel and generous depth of field, keeping both Su-ri-young’s raised profile and the elevated CCTV housing clearly resolved. Let the exterior wall’s vertical plane dominate the composition, while its lines and the open negative space make the lower-left-to-upper-right eyeline diagonal unmistakable. Neutral early-morning light falls evenly across her face, clothing, backpack, wall texture, and camera housing, with subdued contrast and precise material 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): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 밖, 건물 외벽에 매달린 CCTV 카메라를 올려다보는 수리영의 측면.\n\nLOCATION (lock): Outside the supermarket, beneath a CCTV camera mounted high on the building's exterior wall. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From the prescribed chest-height side axis, hold a medium profile of 수리영 in the lower-left portion while the wall-mounted CCTV camera sits above and across her eyeline in the upper-right. Freeze the tilt at its upper extent so the diagonal between her raised gaze and the camera is physically clear before the next downward move.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the lower-left of the frame, midground, looks toward wall-mounted CCTV camera; wall-mounted CCTV camera in the upper-right of the frame, background.\n- KEY BACKGROUND ELEMENTS: 외벽 CCTV 카메라 (Mounted on the market exterior wall) — Its camera-facing housing and lens direction are visible above 수리영's sightline; used as The surveillance camera forms the upper endpoint of 수리영's eyeline and the evidentiary focus of the shot; 마트 외벽 (Supports the CCTV camera); used as The wall provides the vertical plane connecting 수리영 and the elevated camera.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Early-morning exterior light remains neutral and subdued, with controlled contrast linking the figure to the surveillance camera.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same morning light, exterior wall materials, entrance facade, and quiet storefront atmosphere from the reference. Exclude the parked bicycle and the interior aisles if they fall outside this upward view, and show only the woman and the wall-mounted security camera.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s bicycle remains parked outside the mart while she checks the exterior CCTV position, 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 prescribed medium profile with a subtly compressed lens feel and selective depth, holding crisp attention on Su-ri-young’s eye, cheek, and upward-tilted posture while the upper-right CCTV remains distinctly recognizable just beyond the primary focus plane. Use the wall as a quiet, minimally textured field so the visual tension travels directly along her gaze to the surveillance camera. Shape the subdued early-morning light softly across her profile, with gentle falloff toward the elevated fixture and controlled, 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." }, "totals": { "A": 4, "B": 7 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "지정된 미디엄 샷보다 프레임이 너무 넓으며, 지시사항과 달리 불필요한 주변 환경과 자전거가 노출되었습니다." }, { "label": "B", "score": 7, "verdict_ko": "요구된 미디엄 샷 프레이밍과 인물-CCTV 간의 구도를 정확히 구현했으며, 기준 이미지의 디테일을 잘 유지했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S19sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "텍스트 생성을 금지하는 지시와 달리 좌측 상단 간판과 유리문에 글자가 포함되어 있습니다.", "fix_en": "Remove all text from the top-left blue sign and the glass doors, leaving the surfaces blank.", "unfixable": false }, { "issue_ko": "캐릭터 레퍼런스와 이전 장면에서는 머리카락이 귀 뒤로 넘겨져 있으나, 여기서는 머리카락이 귀와 뺨을 덮고 있습니다.", "fix_en": "Tuck the hair on the side of her head behind her ear to match the reference.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove all text from the top-left blue sign and the glass doors, leaving the surfaces blank.\n- Tuck the hair on the side of her head behind her ear to match the 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.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "요구된 앵글과 프레이밍을 완벽하게 구현했으며, 레퍼런스에 있는 장소의 디테일과 캐릭터의 외형 및 소품을 충실히 유지했습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍과 캐릭터의 모습은 좋으나, 레퍼런스 이미지의 유리문에는 존재하지 않는 마스크 포스터(텍스트 포함)를 임의로 생성하여 장소의 일관성을 위반했습니다(Hard Violation)." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "요구된 앵글과 프레이밍을 완벽하게 구현했으며, 레퍼런스에 있는 장소의 디테일과 캐릭터의 외형 및 소품을 충실히 유지했습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍과 캐릭터의 모습은 좋으나, 레퍼런스 이미지의 유리문에는 존재하지 않는 마스크 포스터(텍스트 포함)를 임의로 생성하여 장소의 일관성을 위반했습니다(Hard Violation)." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "참조 이미지에 확립된 마트 간판의 텍스트를 유지하여 장소의 연속성을 더 잘 지켰으며, 지시된 카메라 앵글, 수리영의 측면 프로필, CCTV와의 대각선 구도를 완벽하게 구현했습니다." }, { "label": "A", "score": 8, "verdict_ko": "요구된 프레이밍과 인물 및 배경 요소의 배치 등 전반적인 지시사항을 잘 따랐으나, 참조 이미지에 명확히 존재하는 전면 간판의 텍스트를 임의로 지워버려 장소 일관성에서 약간 감점되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "참조 이미지에 확립된 마트 간판의 텍스트를 유지하여 장소의 연속성을 더 잘 지켰으며, 지시된 카메라 앵글, 수리영의 측면 프로필, CCTV와의 대각선 구도를 완벽하게 구현했습니다." }, { "label": "B", "score": 8, "verdict_ko": "요구된 프레이밍과 인물 및 배경 요소의 배치 등 전반적인 지시사항을 잘 따랐으나, 참조 이미지에 명확히 존재하는 전면 간판의 텍스트를 임의로 지워버려 장소 일관성에서 약간 감점되었습니다." } ] }, "combined": { "totals": { "A": 19, "B": 12 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S19sh1" }, "lane_policy": "ab_select_bypass:prev" }, "S20sh1::bgfirst_bg": { "input_fingerprint": "48753239ea4c12d3", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 마트 총무과 모니터 앞, 어두운 화면을 뚫어지게 주시하는 수리영의 초조한 얼굴 클로즈업.\n\nLOCATION (lock): Inside the supermarket's administrative security office, directly in front of the CCTV monitor under office lighting.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across from 수리영, hold a close three-quarter view that isolates her face while a thin, oblique edge of the monitor remains just outside the central axis. Her eyes track the unseen dark footage past the camera, and the static frame lets anxiety accumulate without a push-in.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (Replaying the section that is too dark to identify) — The display face is turned toward 수리영 and is only partly visible to camera at an oblique edge, showing the unusually dark footage; used as An oblique sliver at the frame edge identifies the source of her fixed attention without displacing her face as the focal point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior ambient light is kept low-key and neutral, allowing the monitor-side contrast to deepen the tension without assigning an unsupported screen color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 마트 총무과 모니터 앞, 어두운 화면을 뚫어지게 주시하는 수리영의 초조한 얼굴 클로즈업.\n\nLOCATION (lock): Inside the supermarket's administrative security office, directly in front of the CCTV monitor under office lighting.\n\nTIME OF DAY (lock): early morning.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across from 수리영, hold a close three-quarter view that isolates her face while a thin, oblique edge of the monitor remains just outside the central axis. Her eyes track the unseen dark footage past the camera, and the static frame lets anxiety accumulate without a push-in.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (Replaying the section that is too dark to identify) — The display face is turned toward 수리영 and is only partly visible to camera at an oblique edge, showing the unusually dark footage; used as An oblique sliver at the frame edge identifies the source of her fixed attention without displacing her face as the focal point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior ambient light is kept low-key and neutral, allowing the monitor-side contrast to deepen the tension without assigning an unsupported screen color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S20sh1__bgfirst_bg.png", "asset_id": "6b56d159-ea8e-419f-a151-71c6463f7fda", "input_asset_ids": [ "6c7d5191-3df7-4880-8cdb-2cee3ebd614a", "3795dc1d-dae4-433f-a34a-691f23ef9290" ] }, "S20sh1": { "input_fingerprint": "b814ff85fe9eccd5", "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.\n\nSHOT TEXT (authoritative, Korean): 마트 총무과 모니터 앞, 어두운 화면을 뚫어지게 주시하는 수리영의 초조한 얼굴 클로즈업.\n\nLOCATION (lock): Inside the supermarket's administrative security office, directly in front of the CCTV monitor under office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across from 수리영, hold a close three-quarter view that isolates her face while a thin, oblique edge of the monitor remains just outside the central axis. Her eyes track the unseen dark footage past the camera, and the static frame lets anxiety accumulate without a push-in.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (Replaying the section that is too dark to identify) — The display face is turned toward 수리영 and is only partly visible to camera at an oblique edge, showing the unusually dark footage; used as An oblique sliver at the frame edge identifies the source of her fixed attention without displacing her face as the focal point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior ambient light is kept low-key and neutral, allowing the monitor-side contrast to deepen the tension without assigning an unsupported screen color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young continues carrying her doll-adorned backpack while she watches the obscured CCTV footage.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 총무과 모니터 앞, 어두운 화면을 뚫어지게 주시하는 수리영의 초조한 얼굴 클로즈업.\n\nLOCATION (lock): Inside the supermarket's administrative security office, directly in front of the CCTV monitor under office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across from 수리영, hold a close three-quarter view that isolates her face while a thin, oblique edge of the monitor remains just outside the central axis. Her eyes track the unseen dark footage past the camera, and the static frame lets anxiety accumulate without a push-in.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (Replaying the section that is too dark to identify) — The display face is turned toward 수리영 and is only partly visible to camera at an oblique edge, showing the unusually dark footage; used as An oblique sliver at the frame edge identifies the source of her fixed attention without displacing her face as the focal point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior ambient light is kept low-key and neutral, allowing the monitor-side contrast to deepen the tension without assigning an unsupported screen color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young continues carrying her doll-adorned backpack while she watches the obscured CCTV footage.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): early morning.\n\nSHOT TEXT (authoritative, Korean): 마트 총무과 모니터 앞, 어두운 화면을 뚫어지게 주시하는 수리영의 초조한 얼굴 클로즈업.\n\nLOCATION (lock): Inside the supermarket's administrative security office, directly in front of the CCTV monitor under office lighting. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At seated eye level directly across from 수리영, hold a close three-quarter view that isolates her face while a thin, oblique edge of the monitor remains just outside the central axis. Her eyes track the unseen dark footage past the camera, and the static frame lets anxiety accumulate without a push-in.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (Replaying the section that is too dark to identify) — The display face is turned toward 수리영 and is only partly visible to camera at an oblique edge, showing the unusually dark footage; used as An oblique sliver at the frame edge identifies the source of her fixed attention without displacing her face as the focal point.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Interior ambient light is kept low-key and neutral, allowing the monitor-side contrast to deepen the tension without assigning an unsupported screen color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young continues carrying her doll-adorned backpack while she watches the obscured CCTV footage.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S20sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S20sh1.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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L10B01.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" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 로케이션의 공간 구조(책상, 창문, 캐비닛의 위치)를 정확히 반영했으며, 인물이 백팩을 메고 있는 상태와 카메라 앵글 지침을 잘 따랐으나 클로즈업 프레이밍이 다소 넓습니다." }, { "label": "B", "score": 5, "verdict_ko": "로케이션 레퍼런스의 공간 구조(벽면 모니터와 책상의 위치)를 왜곡하여 배치했으며, 인물이 백팩을 메고 있지 않고 의자에 걸쳐둔 점이 지침에 어긋납니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 로케이션의 공간 구조(책상, 창문, 캐비닛의 위치)를 정확히 반영했으며, 인물이 백팩을 메고 있는 상태와 카메라 앵글 지침을 잘 따랐으나 클로즈업 프레이밍이 다소 넓습니다." }, { "label": "B", "score": 5, "verdict_ko": "로케이션 레퍼런스의 공간 구조(벽면 모니터와 책상의 위치)를 왜곡하여 배치했으며, 인물이 백팩을 메고 있지 않고 의자에 걸쳐둔 점이 지침에 어긋납니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "지정된 정면 카메라 앵글과 화면 가장자리에 모니터 일부를 걸치게 하는 프레이밍 지침을 완벽하게 준수했으며, 백팩과 인형의 상태도 정확히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "카메라가 인물의 정면이 아닌 측면에 위치하여 명시된 구도 지침을 어겼으며, 프레이밍도 클로즈업보다 넓게 잡혔습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "지정된 정면 카메라 앵글과 화면 가장자리에 모니터 일부를 걸치게 하는 프레이밍 지침을 완벽하게 준수했으며, 백팩과 인형의 상태도 정확히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "카메라가 인물의 정면이 아닌 측면에 위치하여 명시된 구도 지침을 어겼으며, 프레이밍도 클로즈업보다 넓게 잡혔습니다." } ] }, "combined": { "totals": { "A": 17, "B": 9 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 17, "B": 9 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "지정된 로케이션의 공간 구조(책상, 창문, 캐비닛의 위치)를 정확히 반영했으며, 인물이 백팩을 메고 있는 상태와 카메라 앵글 지침을 잘 따랐으나 클로즈업 프레이밍이 다소 넓습니다." }, { "label": "B", "score": 5, "verdict_ko": "로케이션 레퍼런스의 공간 구조(벽면 모니터와 책상의 위치)를 왜곡하여 배치했으며, 인물이 백팩을 메고 있지 않고 의자에 걸쳐둔 점이 지침에 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L10B01.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물이 정면을 보고 있어, 오른쪽 모니터를 주시하는 3/4 측면 포즈를 지시한 레이아웃 스케치와 일치하지 않습니다.", "fix_en": "Rotate the character's body and head into a three-quarter profile facing right, with her eyes tracking the monitor, matching the layout sketch exactly.", "unfixable": false }, { "issue_ko": "프롬프트의 클로즈업 지시 및 스케치에 그려진 인물 크기와 달리, 인물이 화면에 너무 작게(미디엄 샷 크기) 배치되었습니다.", "fix_en": "The framing is too wide and fails to be a close-up; achieving the correct scale requires moving the camera closer, which is restricted.", "unfixable": true } ] }, "composition_critique": { "issues": [ { "fix_en": "Crop tighter around Su-ri-young’s face and shoulders to create a close-up while retaining only a narrow monitor edge at frame right.", "issue_ko": "현재 허리 위까지 넓게 보여 얼굴을 고립시키는 클로즈업보다 프레이밍이 너무 넓고 인물의 화면 크기가 작다.", "unfixable": false }, { "fix_en": "Shift Su-ri-young slightly left so her face is centered in the frame, preserving her scale and gaze direction.", "issue_ko": "수리영의 얼굴이 계약의 중간 중앙보다 오른쪽에 치우쳐 있다.", "unfixable": false }, { "fix_en": "Reduce the visible width of the foreground monitor to a thin oblique sliver at the right frame edge.", "issue_ko": "오른쪽 전경 모니터가 얇은 비스듬한 가장자리가 아니라 넓은 검은 영역으로 보여 얼굴과 시각적 비중을 경쟁한다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Rotate the character's body and head into a three-quarter profile facing right, with her eyes tracking the monitor, matching the layout sketch exactly.\n- Crop tighter around Su-ri-young’s face and shoulders to create a close-up while retaining only a narrow monitor edge at frame right.\n- Shift Su-ri-young slightly left so her face is centered in the frame, preserving her scale and gaze direction.\n- Reduce the visible width of the foreground monitor to a thin oblique sliver at the right frame edge.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "레이아웃 스케치와 달리 얼굴이 너무 정면을 향하고 있어 측면 시선과 자세 지침을 어겼습니다." }, { "label": "B", "score": 8, "verdict_ko": "레이아웃 스케치에 제시된 3/4 측면 얼굴 각도와 모니터 가장자리를 향한 시선을 정확히 구현했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "레이아웃 스케치와 달리 얼굴이 너무 정면을 향하고 있어 측면 시선과 자세 지침을 어겼습니다." }, { "label": "B", "score": 8, "verdict_ko": "레이아웃 스케치에 제시된 3/4 측면 얼굴 각도와 모니터 가장자리를 향한 시선을 정확히 구현했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "레이아웃 스케치와 프롬프트가 요구하는 3/4 측면 구도를 정확히 구현하였으며, 캐릭터와 배경 요소도 충실히 반영되었습니다." }, { "label": "B", "score": 6, "verdict_ko": "캐릭터와 배경은 잘 반영되었으나, 요구된 3/4 측면 구도 대신 거의 정면을 향하고 있어 레이아웃 지침을 어겼습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "레이아웃 스케치와 프롬프트가 요구하는 3/4 측면 구도를 정확히 구현하였으며, 캐릭터와 배경 요소도 충실히 반영되었습니다." }, { "label": "A", "score": 6, "verdict_ko": "캐릭터와 배경은 잘 반영되었으나, 요구된 3/4 측면 구도 대신 거의 정면을 향하고 있어 레이아웃 지침을 어겼습니다." } ] }, "combined": { "totals": { "A": 11, "B": 17 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S20sh1__bgfirst_bg.png", "bg_asset_id": "6b56d159-ea8e-419f-a151-71c6463f7fda", "bg_record_key": "S20sh1::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S20sh4::variants": { "author_fp": "902515a358a69d74", "author": { "variants": [ { "approach_ko": "얕은 심도와 완만한 망원 압축으로 혜수의 단호한 시선에 집중하는 긴장감 있는 접근.", "prompt_en": "Use a restrained telephoto feel with shallow depth, gently compressing the eyeline pair and isolating 혜수’s face in crisp focus. Let the foreground shoulder and partial profile fall softly out of focus, while the oblique monitor edge and its dark display remain dimly legible behind them. Shape the neutral office ambience with subtle side modeling across her mature features, preserving controlled shadow detail and a muted, low-key finish." }, { "approach_ko": "더 넓은 공간감과 깊은 초점층으로 인물 사이의 대치와 CCTV 맥락을 함께 살리는 접근.", "prompt_en": "Use a close, natural perspective with stronger depth separation through layered blocking rather than heavy optical blur. Keep 혜수’s firm gaze and the foreground partial profile clearly readable across the shared eyeline, while allowing the oblique CCTV monitor to retain recognizable texture and the unusually dark reviewed segment without competing for attention. Favor even early-morning office ambience with restrained contrast, subtle practical falloff, and an observational sense of immediate confrontation." } ] }, "reused": false }, "S20sh4": { "input_fingerprint": "5371c4d0c3338af4", "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.\n\nSHOT TEXT (authoritative, Korean): 한국인 남성 보안 직원을 향해 단호한 눈빛을 띠는 혜수의 얼굴.\n\nLOCATION (lock): Inside the supermarket's administrative security office, beside the CCTV workstation and security employee. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: End the lateral flow at 혜수's eye level in a close three-quarter composition, placing her firm face near center-left and the 한국인 남성 보안 직원's shoulder and partial profile at the lower-right foreground. Keep their shared eyeline unobstructed so her request to take the CCTV recording reads as a decision directed at him, not as a frontal address to the audience.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 혜수 in the middle-left of the frame, midground, looks toward security employee; 한국인 남성 보안 직원 in the lower-right of the frame, foreground, looks toward 혜수.\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (Showing the unusually dark CCTV section) — Its display face is seen obliquely behind the eyeline pair, with the dark reviewed segment still present; used as The monitor edge behind the pair preserves continuity from the CCTV review while remaining subordinate to 혜수's expression.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral interior ambience and restrained low-key contrast sharpen 혜수's resolve while keeping the palette muted.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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.\n\nSHOT TEXT (authoritative, Korean): 한국인 남성 보안 직원을 향해 단호한 눈빛을 띠는 혜수의 얼굴.\n\nLOCATION (lock): Inside the supermarket's administrative security office, beside the CCTV workstation and security employee. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: End the lateral flow at 혜수's eye level in a close three-quarter composition, placing her firm face near center-left and the 한국인 남성 보안 직원's shoulder and partial profile at the lower-right foreground. Keep their shared eyeline unobstructed so her request to take the CCTV recording reads as a decision directed at him, not as a frontal address to the audience.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 혜수 in the middle-left of the frame, midground, looks toward security employee; 한국인 남성 보안 직원 in the lower-right of the frame, foreground, looks toward 혜수.\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (Showing the unusually dark CCTV section) — Its display face is seen obliquely behind the eyeline pair, with the dark reviewed segment still present; used as The monitor edge behind the pair preserves continuity from the CCTV review while remaining subordinate to 혜수's expression.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral interior ambience and restrained low-key contrast sharpen 혜수's resolve while keeping the palette muted.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 telephoto feel with shallow depth, gently compressing the eyeline pair and isolating 혜수’s face in crisp focus. Let the foreground shoulder and partial profile fall softly out of focus, while the oblique monitor edge and its dark display remain dimly legible behind them. Shape the neutral office ambience with subtle side modeling across her mature features, preserving controlled shadow detail and a muted, low-key finish.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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.\n\nSHOT TEXT (authoritative, Korean): 한국인 남성 보안 직원을 향해 단호한 눈빛을 띠는 혜수의 얼굴.\n\nLOCATION (lock): Inside the supermarket's administrative security office, beside the CCTV workstation and security employee. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: End the lateral flow at 혜수's eye level in a close three-quarter composition, placing her firm face near center-left and the 한국인 남성 보안 직원's shoulder and partial profile at the lower-right foreground. Keep their shared eyeline unobstructed so her request to take the CCTV recording reads as a decision directed at him, not as a frontal address to the audience.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: 혜수 in the middle-left of the frame, midground, looks toward security employee; 한국인 남성 보안 직원 in the lower-right of the frame, foreground, looks toward 혜수.\n- KEY BACKGROUND ELEMENTS: CCTV 모니터 (Showing the unusually dark CCTV section) — Its display face is seen obliquely behind the eyeline pair, with the dark reviewed segment still present; used as The monitor edge behind the pair preserves continuity from the CCTV review while remaining subordinate to 혜수's expression.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral interior ambience and restrained low-key contrast sharpen 혜수's resolve while keeping the palette muted.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\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 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close, natural perspective with stronger depth separation through layered blocking rather than heavy optical blur. Keep 혜수’s firm gaze and the foreground partial profile clearly readable across the shared eyeline, while allowing the oblique CCTV monitor to retain recognizable texture and the unusually dark reviewed segment without competing for attention. Favor even early-morning office ambience with restrained contrast, subtle practical falloff, and an observational sense of immediate confrontation.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 8, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "제공된 혜수의 캐릭터 레퍼런스를 정확히 반영했으며, 요구된 클로즈업 구도와 인물 배치를 충실하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "혜수 캐릭터 레퍼런스 대신 이전 컷의 인물을 잘못 등장시켰고, 프레임이 지정된 클로즈업보다 넓으며 제복에 오타('POLCE')가 발생했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S20sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "이전 샷과의 연속성을 유지해야 한다는 지시와 달리, 혜수가 이전 샷의 회색 후드티가 아닌 경찰 제복을 입고 있습니다.", "fix_en": "Replace Hye-su's dark navy police jacket with the light grey hooded sweatshirt she wears in the previous shot still.", "unfixable": false }, { "issue_ko": "보안 직원의 어깨 견장과 혜수의 어깨에 의미를 알 수 없는 가짜 문자와 기호가 있어 '텍스트 없음' 규칙을 위반합니다.", "fix_en": "Remove the white illegible text and symbols from the shoulder epaulettes of both characters, leaving them as plain dark fabric.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Replace Hye-su's dark navy police jacket with the light grey hooded sweatshirt she wears in the previous shot still.\n- Remove the white illegible text and symbols from the shoulder epaulettes of both characters, leaving them as plain dark fabric.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "이전 컷에서 확립된 혜수의 의상(회색 후드티) 연속성을 엄격하게 유지하며, 제시된 구도와 인물 배치를 정확하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "캐릭터 레퍼런스의 경찰 제복을 입혀 이전 컷에서 확립된 의상 연속성을 심각하게 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "이전 컷에서 확립된 혜수의 의상(회색 후드티) 연속성을 엄격하게 유지하며, 제시된 구도와 인물 배치를 정확하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "캐릭터 레퍼런스의 경찰 제복을 입혀 이전 컷에서 확립된 의상 연속성을 심각하게 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "이전 샷의 연속성(회색 후드티 의상)을 정확히 유지했으며 명시된 구도와 캐릭터 배치도 잘 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "인물 의상을 이전 샷의 후드티가 아닌 캐릭터 레퍼런스의 제복으로 임의 변경하여 연속성 규칙을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "이전 샷의 연속성(회색 후드티 의상)을 정확히 유지했으며 명시된 구도와 캐릭터 배치도 잘 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "인물 의상을 이전 샷의 후드티가 아닌 캐릭터 레퍼런스의 제복으로 임의 변경하여 연속성 규칙을 위반했습니다." } ] }, "combined": { "totals": { "A": 7, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S20sh1" } }, "S21sh6::variants": { "author_fp": "1b15bdb5c3067e26", "author": { "variants": [ { "approach_ko": "빈 의자를 흐릿한 전경 앵커로 두고 모니터 결과창에 시선을 강하게 집중시키는 얕은 심도 접근.", "prompt_en": "Use a compressed, intimate close-up from monitor height, letting the small lower-edge portion of the vacant seat fall softly out of focus while the monitor frame and its central result remain crisply resolved. Keep the display nearly square to camera and hold the surrounding workstation in restrained midtone shadow, with neutral daytime ambience and precise screen exposure creating a quiet, urgent point of attention." }, { "approach_ko": "빈 자리와 모니터를 함께 또렷하게 유지해 사람이 없는 업무 공간의 정적과 정보의 돌발성을 강조하는 깊은 심도 접근.", "prompt_en": "Photograph the close-up with greater depth and a firmer spatial relationship between the vacant seat edge and the full monitor frame, preserving both as clearly physical parts of the unattended workstation. Use the slight off-axis position behind the chair to create subtle converging depth toward the centered display, with even, restrained daytime office light and a clinical tonal balance that makes the sudden result feel stark rather than sensational." } ] }, "reused": false }, "S21sh6": { "input_fingerprint": "9febb39819d7f23b", "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): 김 형사의 빈 자리, 모니터 화면 한가운데 위치 추적 결과창이 선명하게 떠오른 찰나.\n\nLOCATION (lock): At an unattended detective's desk inside the police office, focused on the lit computer monitor displaying a location result. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Begin behind and slightly beside 김 형사's vacant position at monitor height, with the empty seat forming a partial lower-edge anchor and the location-tracking result centered on the display. Start the prescribed dolly inward without yet losing the monitor frame, making the newly appeared result unmistakable while preserving the unattended workstation context.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: location-tracking result on monitor in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 김 형사의 모니터 (Showing 강민숙's phone location-tracking result) — The display face is square enough to camera for the location result, sea map, coordinates, timestamp, and offline status to be legible at center; used as The display is the focal surface carrying the newly received tracking information; 김 형사의 빈 자리 (Vacant) — The rear and side of the empty seat occupy a small lower-frame area in front of the monitor; used as The vacant position confirms that the information has appeared while 김 형사 is away.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daytime office ambience remains neutral and restrained, with the monitor content receiving clear exposure for legibility.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook’s phone-location result remains displayed on Detective Kim’s monitor, showing an offshore coordinate and an offline status. Su-ri-young still has her backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 김 형사의 빈 자리, 모니터 화면 한가운데 위치 추적 결과창이 선명하게 떠오른 찰나.\n\nLOCATION (lock): At an unattended detective's desk inside the police office, focused on the lit computer monitor displaying a location result. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Begin behind and slightly beside 김 형사's vacant position at monitor height, with the empty seat forming a partial lower-edge anchor and the location-tracking result centered on the display. Start the prescribed dolly inward without yet losing the monitor frame, making the newly appeared result unmistakable while preserving the unattended workstation context.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: location-tracking result on monitor in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 김 형사의 모니터 (Showing 강민숙's phone location-tracking result) — The display face is square enough to camera for the location result, sea map, coordinates, timestamp, and offline status to be legible at center; used as The display is the focal surface carrying the newly received tracking information; 김 형사의 빈 자리 (Vacant) — The rear and side of the empty seat occupy a small lower-frame area in front of the monitor; used as The vacant position confirms that the information has appeared while 김 형사 is away.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daytime office ambience remains neutral and restrained, with the monitor content receiving clear exposure for legibility.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook’s phone-location result remains displayed on Detective Kim’s monitor, showing an offshore coordinate and an offline status. Su-ri-young still has her backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed, intimate close-up from monitor height, letting the small lower-edge portion of the vacant seat fall softly out of focus while the monitor frame and its central result remain crisply resolved. Keep the display nearly square to camera and hold the surrounding workstation in restrained midtone shadow, with neutral daytime ambience and precise screen exposure creating a quiet, urgent point of attention.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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): 김 형사의 빈 자리, 모니터 화면 한가운데 위치 추적 결과창이 선명하게 떠오른 찰나.\n\nLOCATION (lock): At an unattended detective's desk inside the police office, focused on the lit computer monitor displaying a location result. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Begin behind and slightly beside 김 형사's vacant position at monitor height, with the empty seat forming a partial lower-edge anchor and the location-tracking result centered on the display. Start the prescribed dolly inward without yet losing the monitor frame, making the newly appeared result unmistakable while preserving the unattended workstation context.\n- FRAMING SCALE: close-up\n- FRAME LAYOUT: location-tracking result on monitor in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: 김 형사의 모니터 (Showing 강민숙's phone location-tracking result) — The display face is square enough to camera for the location result, sea map, coordinates, timestamp, and offline status to be legible at center; used as The display is the focal surface carrying the newly received tracking information; 김 형사의 빈 자리 (Vacant) — The rear and side of the empty seat occupy a small lower-frame area in front of the monitor; used as The vacant position confirms that the information has appeared while 김 형사 is away.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daytime office ambience remains neutral and restrained, with the monitor content receiving clear exposure for legibility.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Min-sook’s phone-location result remains displayed on Detective Kim’s monitor, showing an offshore coordinate and an offline status. Su-ri-young still has her backpack with her.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the close-up with greater depth and a firmer spatial relationship between the vacant seat edge and the full monitor frame, preserving both as clearly physical parts of the unattended workstation. Use the slight off-axis position behind the chair to create subtle converging depth toward the centered display, with even, restrained daytime office light and a clinical tonal balance that makes the sudden result feel stark rather than sensational.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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 대로 빈 의자를 하단 앵커로 삼은 구도 속에서 모니터 중앙에 위치 추적 결과창과 오프라인 상태를 명확하게 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "전반적인 구도와 배경은 일치하나, 핵심인 중앙의 선명한 팝업 결과창과 오프라인 상태 표시가 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L13B01.png" } ], "critique": { "issues": [ { "issue_ko": "데스크 램프의 금속 기둥이 테이프 디스펜서 위로 직접 연결되어 두 사물이 비현실적으로 융합되어 있습니다.", "fix_en": "Redraw the base of the desk lamp as a standard round stand, separating it completely from the tape dispenser.", "unfixable": false }, { "issue_ko": "모니터 화면 중앙의 정보 창이 자연스러운 소프트웨어 UI가 아니라 이미지 위에 인위적으로 덧붙인 그래픽 오버레이나 텍스트 캡션처럼 보입니다.", "fix_en": "Redesign the central text box to look like a natural diegetic UI panel integrated into the map software, removing the artificial pasted-on overlay appearance.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "issue_ko": "프레임이 너무 넓어 사무실 주변이 크게 포함되고 모니터가 클로즈업 크기보다 작게 보인다.", "fix_en": "Crop tightly around the monitor while retaining its full frame and keeping the tracking result centered.", "unfixable": false }, { "issue_ko": "빈 의자가 하단의 작은 앵커가 아니라 화면 하단을 넓고 크게 차지한다.", "fix_en": "Trim the lower frame so only a small, softly blurred upper edge of the vacant seat remains visible.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw the base of the desk lamp as a standard round stand, separating it completely from the tape dispenser.\n- Redesign the central text box to look like a natural diegetic UI panel integrated into the map software, removing the artificial pasted-on overlay appearance.\n- Crop tightly around the monitor while retaining its full frame and keeping the tracking result centered.\n- Trim the lower frame so only a small, softly blurred upper edge of the vacant seat remains visible.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 앵글과 하단 앵커 구도를 잘 구현했으며, 모니터 화면의 위치 추적 정보 텍스트가 매우 정확합니다." }, { "label": "B", "score": 5, "verdict_ko": "구도와 톤은 적절하나, 모니터 화면 내 텍스트에 철자 오류('위치' 오타, 영문 표기 오류)가 발생했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 앵글과 하단 앵커 구도를 잘 구현했으며, 모니터 화면의 위치 추적 정보 텍스트가 매우 정확합니다." }, { "label": "B", "score": 5, "verdict_ko": "구도와 톤은 적절하나, 모니터 화면 내 텍스트에 철자 오류('위치' 오타, 영문 표기 오류)가 발생했습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 화면 내 텍스트('강민숙', 오프라인 상태, 좌표 등)를 정확하게 생성하였으며, 빈 의자가 하단을 차지하는 구도와 심도 표현도 훌륭합니다." }, { "label": "A", "score": 5, "verdict_ko": "구도와 배경은 적절하나, 모니터 화면의 대상 이름('강민숙'이 아닌 '김민숙')과 영문 오타 등 텍스트 정확도에서 오류가 있습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 화면 내 텍스트('강민숙', 오프라인 상태, 좌표 등)를 정확하게 생성하였으며, 빈 의자가 하단을 차지하는 구도와 심도 표현도 훌륭합니다." }, { "label": "B", "score": 5, "verdict_ko": "구도와 배경은 적절하나, 모니터 화면의 대상 이름('강민숙'이 아닌 '김민숙')과 영문 오타 등 텍스트 정확도에서 오류가 있습니다." } ] }, "combined": { "totals": { "A": 14, "B": 10 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L13B01", "B": "L13B02", "C": "L13B03", "D": "L13B04" }, "assigned": "L13B01", "choice": "A", "confident": true, "reason_ko": "\"모니터 화면 한가운데 위치 추적 결과창이 선명하게 떠오른 찰나\"라는 묘사에 맞게 모니터 화면이 클로즈업된 A가 적합함.", "chosen": "L13B01" } }, "S21sh9::variants": { "author_fp": "7bcf68a0f5ac66ee", "author": { "variants": [ { "approach_ko": "얕은 심도로 몸의 비틀림과 손의 긴장을 밀착해 포착하는 심리적 접근.", "prompt_en": "Use a restrained, slightly compressed lens feel and shallow depth of field, making the twisted shoulder line, backward glance, and tense hands the immediate visual emphasis. Hold the background just soft enough to feel unobserved yet unmistakably vacant, with subdued daytime illumination shaping her against the darker room. Preserve natural fabric creases, skin texture, worn paper, and the cool metallic response of the handle." }, { "approach_ko": "깊은 심도와 문틀의 기하학으로 비밀스러운 이탈과 텅 빈 공간을 강조하는 환경적 접근.", "prompt_en": "Use a wider, rectilinear lens feel while retaining the prescribed medium scale, with deep focus carrying clearly from her back and hands into the receding work area. Let the doorway geometry create a firm visual boundary and use the upper-left emptiness as deliberate negative space, making her departure feel exposed and irreversible. Keep the neutral daylight controlled and contrasty, with crisp spatial separation and realistic surface wear throughout." } ] }, "reused": false }, "S21sh9": { "input_fingerprint": "b6d1d3ea741c1cd5", "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): 구겨진 메모지를 움켜쥔 채 문손잡이를 잡고 어깨 너머로 뒤를 돌아보는 수리영의 뒷모습.\n\nLOCATION (lock): At the police office doorway leading into the station corridor, with the occupied workroom behind her in daylight hours. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track directly behind 수리영 at shoulder height until she reaches the office door, keeping her back in the central midground and the hand holding the crumpled note visible near her side. As her other hand grips the handle, catch the torso twist and over-shoulder glance back into the emptied office, with the doorway forming the exit boundary ahead.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, looks toward empty police office behind her; empty police office behind her in the upper-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 사무실 문 (수리영 grips the handle as she begins to leave) — The interior-facing side and handle are visible ahead of the camera behind 수리영; used as The door and handle provide the exit anchor for both hands and define the threshold of her secret departure; 경찰서 사무실 내부 (Empty after the detectives have left) — The desks and work area recede behind her turning shoulder; used as The rear office space receives her backward glance and confirms that no one remains to observe her.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral daytime office ambience is held in moody contrast, separating 수리영 from the unoccupied room behind her.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young has copied the offshore coordinates onto a note, folded it, and put it in her pocket; her backpack remains with her as she leaves.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 구겨진 메모지를 움켜쥔 채 문손잡이를 잡고 어깨 너머로 뒤를 돌아보는 수리영의 뒷모습.\n\nLOCATION (lock): At the police office doorway leading into the station corridor, with the occupied workroom behind her in daylight hours. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track directly behind 수리영 at shoulder height until she reaches the office door, keeping her back in the central midground and the hand holding the crumpled note visible near her side. As her other hand grips the handle, catch the torso twist and over-shoulder glance back into the emptied office, with the doorway forming the exit boundary ahead.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, looks toward empty police office behind her; empty police office behind her in the upper-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 사무실 문 (수리영 grips the handle as she begins to leave) — The interior-facing side and handle are visible ahead of the camera behind 수리영; used as The door and handle provide the exit anchor for both hands and define the threshold of her secret departure; 경찰서 사무실 내부 (Empty after the detectives have left) — The desks and work area recede behind her turning shoulder; used as The rear office space receives her backward glance and confirms that no one remains to observe her.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral daytime office ambience is held in moody contrast, separating 수리영 from the unoccupied room behind her.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young has copied the offshore coordinates onto a note, folded it, and put it in her pocket; her backpack remains with her as she leaves.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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, slightly compressed lens feel and shallow depth of field, making the twisted shoulder line, backward glance, and tense hands the immediate visual emphasis. Hold the background just soft enough to feel unobserved yet unmistakably vacant, with subdued daytime illumination shaping her against the darker room. Preserve natural fabric creases, skin texture, worn paper, and the cool metallic response of the handle.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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): 구겨진 메모지를 움켜쥔 채 문손잡이를 잡고 어깨 너머로 뒤를 돌아보는 수리영의 뒷모습.\n\nLOCATION (lock): At the police office doorway leading into the station corridor, with the occupied workroom behind her in daylight hours. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track directly behind 수리영 at shoulder height until she reaches the office door, keeping her back in the central midground and the hand holding the crumpled note visible near her side. As her other hand grips the handle, catch the torso twist and over-shoulder glance back into the emptied office, with the doorway forming the exit boundary ahead.\n- FRAMING SCALE: medium shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, looks toward empty police office behind her; empty police office behind her in the upper-left of the frame, background.\n- KEY BACKGROUND ELEMENTS: 사무실 문 (수리영 grips the handle as she begins to leave) — The interior-facing side and handle are visible ahead of the camera behind 수리영; used as The door and handle provide the exit anchor for both hands and define the threshold of her secret departure; 경찰서 사무실 내부 (Empty after the detectives have left) — The desks and work area recede behind her turning shoulder; used as The rear office space receives her backward glance and confirms that no one remains to observe her.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Neutral daytime office ambience is held in moody contrast, separating 수리영 from the unoccupied room behind her.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young has copied the offshore coordinates onto a note, folded it, and put it in her pocket; her backpack remains with her as she leaves.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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, rectilinear lens feel while retaining the prescribed medium scale, with deep focus carrying clearly from her back and hands into the receding work area. Let the doorway geometry create a firm visual boundary and use the upper-left emptiness as deliberate negative space, making her departure feel exposed and irreversible. Keep the neutral daylight controlled and contrasty, with crisp spatial separation and realistic surface wear throughout.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": "프롬프트가 요구한 카메라 위치(등 뒤), 어깨 너머로 뒤돌아보는 포즈, 구겨진 메모지와 문손잡이를 잡은 양손의 디테일을 완벽하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "'뒷모습'을 촬영하라는 핵심 지시와 달리 피사체의 앞/옆모습을 보여주어 카메라 시점과 샷 구성을 심각하게 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S21sh6_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "문이 열린 틈 사이, 문 가장자리 우측에 불필요한 두 번째 은색 문손잡이가 허공에 떠 있습니다.", "fix_en": "Remove the extra silver doorknob floating in the gap to the right of the door edge.", "unfixable": false }, { "issue_ko": "배경의 정수기 위 파란색 안내판에 흰색 문자가 적혀 있어 텍스트 금지 규칙을 위반했습니다.", "fix_en": "Remove all text from the blue sign on the background wall, leaving it blank.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the extra silver doorknob floating in the gap to the right of the door edge.\n- Remove all text from the blue sign on the background wall, leaving it 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.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "지정된 프레이밍과 연출을 잘 구현했으나, 문의 형태(경첩과 손잡이가 같은 면에 위치)에 물리적 오류가 있습니다." }, { "label": "A", "score": 3, "verdict_ko": "B와 동일한 문의 구조적 오류가 있으며, 메모지를 쥔 왼손이 해부학적으로 불가능하게(오른손으로) 렌더링되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "지정된 프레이밍과 연출을 잘 구현했으나, 문의 형태(경첩과 손잡이가 같은 면에 위치)에 물리적 오류가 있습니다." }, { "label": "A", "score": 3, "verdict_ko": "B와 동일한 문의 구조적 오류가 있으며, 메모지를 쥔 왼손이 해부학적으로 불가능하게(오른손으로) 렌더링되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "구겨진 메모지를 쥔 손, 문손잡이를 잡은 자세, 어깨 너머로 돌아보는 시선과 카메라 구도 등 프롬프트의 요구사항을 매우 정확하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "문손잡이를 잡은 손 바로 옆에 또 다른 손잡이가 허공에 생성되어 물리적으로 불가능한 구조(Hard Violation)를 보입니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "구겨진 메모지를 쥔 손, 문손잡이를 잡은 자세, 어깨 너머로 돌아보는 시선과 카메라 구도 등 프롬프트의 요구사항을 매우 정확하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "문손잡이를 잡은 손 바로 옆에 또 다른 손잡이가 허공에 생성되어 물리적으로 불가능한 구조(Hard Violation)를 보입니다." } ] }, "combined": { "totals": { "A": 6, "B": 11 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S21sh6" } }, "groupbg::경찰서 앞 도로": { "input_fingerprint": "0230d7bb6e9d1e6e", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "경찰서 앞 도로", "tags": [ "S22sh1" ] }, "context_sig": "2bbb7f3a1b9aebd7" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: On the road immediately outside the police station, where the parked bicycle pulls away in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n서해안 해안도로: 바다와 인접한 해안가 포장도로로, 자전거가 주행할 수 있는 도로폭을 갖추고 있다. (특징: 아스팔트 도로; 바다와 맞닿은 방파제 구조물; 세차게 쏟아지는 비와 젖은 노면(일부 시점))\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#22. 경찰서 밖 / D\n- 주차된 자전거를 타고 이내 달리는 수리영\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: On the road immediately outside the police station, where the parked bicycle pulls away in daylight.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n서해안 해안도로: 바다와 인접한 해안가 포장도로로, 자전거가 주행할 수 있는 도로폭을 갖추고 있다. (특징: 아스팔트 도로; 바다와 맞닿은 방파제 구조물; 세차게 쏟아지는 비와 젖은 노면(일부 시점))\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#22. 경찰서 밖 / D\n- 주차된 자전거를 타고 이내 달리는 수리영\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_경찰서_앞_도로_0f1b35.png", "asset_id": "bb0e0a7c-4bce-45f2-b5b4-94f916494f88", "input_asset_ids": [ "a85aaaf6-b5db-4f6a-a729-c049d42014a7" ], "origin_tag": "S22sh1", "place_text": "On the road immediately outside the police station, where the parked bicycle pulls away in daylight.", "origin_inputs": { "place_text": "On the road immediately outside the police station, where the parked bicycle pulls away in daylight.", "time_of_day_en": "day", "conti_asset_id": "a85aaaf6-b5db-4f6a-a729-c049d42014a7" } }, "S22sh1::bgfirst_bg": { "input_fingerprint": "19e977a5ff92c81e", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 경찰서 밖 도로, 상체를 웅크린 채 자전거 페달에 체중을 싣는 수리영의 측면 전신.\n\nLOCATION (lock): On the road immediately outside the police station, where the parked bicycle pulls away in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Maintain the prescribed waist-height parallel track at full-body side distance, placing 수리영 and the bicycle across the middle of the frame as they accelerate toward screen right. Hold the profile axis while she folds over the handlebars and transfers her weight into the pedals, leaving forward space in the direction of travel.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen right along the road.\n- KEY BACKGROUND ELEMENTS: 자전거 (In motion under hard pedaling) — Its full side profile faces the tracking camera as it travels toward screen right; used as The bicycle remains fully readable with 수리영 as a single moving silhouette and provides the mechanical line of acceleration; 경찰서 밖 도로 (수리영 rides away along it); used as The road supplies the lateral travel axis beneath the parallel tracking move.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior is rendered in muted neutrals with firm, low-key contrast to sustain urgency.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 경찰서 밖 도로, 상체를 웅크린 채 자전거 페달에 체중을 싣는 수리영의 측면 전신.\n\nLOCATION (lock): On the road immediately outside the police station, where the parked bicycle pulls away in daylight.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Maintain the prescribed waist-height parallel track at full-body side distance, placing 수리영 and the bicycle across the middle of the frame as they accelerate toward screen right. Hold the profile axis while she folds over the handlebars and transfers her weight into the pedals, leaving forward space in the direction of travel.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen right along the road.\n- KEY BACKGROUND ELEMENTS: 자전거 (In motion under hard pedaling) — Its full side profile faces the tracking camera as it travels toward screen right; used as The bicycle remains fully readable with 수리영 as a single moving silhouette and provides the mechanical line of acceleration; 경찰서 밖 도로 (수리영 rides away along it); used as The road supplies the lateral travel axis beneath the parallel tracking move.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior is rendered in muted neutrals with firm, low-key contrast to sustain urgency.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S22sh1__bgfirst_bg.png", "asset_id": "fd3f2282-572e-4d79-83ad-4e4e86c70583", "input_asset_ids": [ "a85aaaf6-b5db-4f6a-a729-c049d42014a7", "bb0e0a7c-4bce-45f2-b5b4-94f916494f88" ] }, "S22sh1": { "input_fingerprint": "32f0941708b5cd22", "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): 경찰서 밖 도로, 상체를 웅크린 채 자전거 페달에 체중을 싣는 수리영의 측면 전신.\n\nLOCATION (lock): On the road immediately outside the police station, where the parked bicycle pulls away in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Maintain the prescribed waist-height parallel track at full-body side distance, placing 수리영 and the bicycle across the middle of the frame as they accelerate toward screen right. Hold the profile axis while she folds over the handlebars and transfers her weight into the pedals, leaving forward space in the direction of travel.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen right along the road.\n- KEY BACKGROUND ELEMENTS: 자전거 (In motion under hard pedaling) — Its full side profile faces the tracking camera as it travels toward screen right; used as The bicycle remains fully readable with 수리영 as a single moving silhouette and provides the mechanical line of acceleration; 경찰서 밖 도로 (수리영 rides away along it); used as The road supplies the lateral travel axis beneath the parallel tracking move.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior is rendered in muted neutrals with firm, low-key contrast to sustain urgency.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young rides away on her bicycle with the folded coordinate note in her pocket and her backpack still 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 경찰서 밖 도로, 상체를 웅크린 채 자전거 페달에 체중을 싣는 수리영의 측면 전신.\n\nLOCATION (lock): On the road immediately outside the police station, where the parked bicycle pulls away in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Maintain the prescribed waist-height parallel track at full-body side distance, placing 수리영 and the bicycle across the middle of the frame as they accelerate toward screen right. Hold the profile axis while she folds over the handlebars and transfers her weight into the pedals, leaving forward space in the direction of travel.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen right along the road.\n- KEY BACKGROUND ELEMENTS: 자전거 (In motion under hard pedaling) — Its full side profile faces the tracking camera as it travels toward screen right; used as The bicycle remains fully readable with 수리영 as a single moving silhouette and provides the mechanical line of acceleration; 경찰서 밖 도로 (수리영 rides away along it); used as The road supplies the lateral travel axis beneath the parallel tracking move.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior is rendered in muted neutrals with firm, low-key contrast to sustain urgency.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young rides away on her bicycle with the folded coordinate note in her pocket and her backpack still 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 경찰서 밖 도로, 상체를 웅크린 채 자전거 페달에 체중을 싣는 수리영의 측면 전신.\n\nLOCATION (lock): On the road immediately outside the police station, where the parked bicycle pulls away in daylight. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Maintain the prescribed waist-height parallel track at full-body side distance, placing 수리영 and the bicycle across the middle of the frame as they accelerate toward screen right. Hold the profile axis while she folds over the handlebars and transfers her weight into the pedals, leaving forward space in the direction of travel.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-center of the frame, midground, moves toward screen right along the road.\n- KEY BACKGROUND ELEMENTS: 자전거 (In motion under hard pedaling) — Its full side profile faces the tracking camera as it travels toward screen right; used as The bicycle remains fully readable with 수리영 as a single moving silhouette and provides the mechanical line of acceleration; 경찰서 밖 도로 (수리영 rides away along it); used as The road supplies the lateral travel axis beneath the parallel tracking move.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Daylight appropriate to the exterior is rendered in muted neutrals with firm, low-key contrast to sustain urgency.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young rides away on her bicycle with the folded coordinate note in her pocket and her backpack still 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S22sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S22sh1.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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_경찰서_앞_도로_0f1b35.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" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 측면 전신 샷과 상체를 웅크린 포즈를 완벽하게 구현했으며, 레퍼런스의 의상(회색 후드, 청바지)과 필수 유지 사항인 배낭을 정확히 묘사했습니다." }, { "label": "B", "score": 6, "verdict_ko": "요구된 프레이밍과 포즈는 따랐으나, 캐릭터의 의상 색상이 레퍼런스와 일치하지 않으며 명시된 배낭이 누락되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 측면 전신 샷과 상체를 웅크린 포즈를 완벽하게 구현했으며, 레퍼런스의 의상(회색 후드, 청바지)과 필수 유지 사항인 배낭을 정확히 묘사했습니다." }, { "label": "B", "score": 6, "verdict_ko": "요구된 프레이밍과 포즈는 따랐으나, 캐릭터의 의상 색상이 레퍼런스와 일치하지 않으며 명시된 배낭이 누락되었습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 6, "verdict_ko": "구도와 배경은 훌륭하나 캐릭터의 의상(회색 후드티, 청바지)이 레퍼런스와 전혀 일치하지 않으며 필수 소지품인 배낭이 누락되었습니다." }, { "label": "B", "score": 10, "verdict_ko": "지시된 측면 전신 구도와 역동적인 자전거 탑승 자세를 잘 구현했으며, 레퍼런스의 의상과 요구된 배낭 착용 상태까지 완벽하게 반영했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 6, "verdict_ko": "구도와 배경은 훌륭하나 캐릭터의 의상(회색 후드티, 청바지)이 레퍼런스와 전혀 일치하지 않으며 필수 소지품인 배낭이 누락되었습니다." }, { "label": "A", "score": 10, "verdict_ko": "지시된 측면 전신 구도와 역동적인 자전거 탑승 자세를 잘 구현했으며, 레퍼런스의 의상과 요구된 배낭 착용 상태까지 완벽하게 반영했습니다." } ] }, "combined": { "totals": { "A": 20, "B": 12 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 } }, "totals": { "A": 20, "B": 12 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 측면 전신 샷과 상체를 웅크린 포즈를 완벽하게 구현했으며, 레퍼런스의 의상(회색 후드, 청바지)과 필수 유지 사항인 배낭을 정확히 묘사했습니다." }, { "label": "B", "score": 6, "verdict_ko": "요구된 프레이밍과 포즈는 따랐으나, 캐릭터의 의상 색상이 레퍼런스와 일치하지 않으며 명시된 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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_경찰서_앞_도로_0f1b35.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "배경 담벼락의 원래 장식이 텍스트가 포함된 거대한 경찰 마크로 임의 변경되었습니다.", "fix_en": "Restore the wall's original small gold emblem, completely removing the added police shield and its text.", "unfixable": false }, { "issue_ko": "자전거 체인링 뒤편에 있어야 할 왼쪽 크랭크 암과 페달이 누락되었습니다.", "fix_en": "Add the missing left crank arm and pedal extending behind the center of the chainring.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Restore the wall's original small gold emblem, completely removing the added police shield and its text.\n- Add the missing left crank arm and pedal extending behind the center of the chainring.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 레이아웃과 배경을 충실히 따랐으며, 상체를 웅크린 채 페달을 밟는 인물의 자세와 자전거의 물리적 구조를 안정적으로 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "전반적인 구도와 인물의 외양은 기준에 부합하나, 자전거의 체인링과 페달 부근에 물리적으로 불가능한 구조적 오류(여분의 부품)가 발생하여 감점되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 레이아웃과 배경을 충실히 따랐으며, 상체를 웅크린 채 페달을 밟는 인물의 자세와 자전거의 물리적 구조를 안정적으로 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "전반적인 구도와 인물의 외양은 기준에 부합하나, 자전거의 체인링과 페달 부근에 물리적으로 불가능한 구조적 오류(여분의 부품)가 발생하여 감점되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 샷의 구도와 인물의 자세를 충실히 구현했으며, 금지된 텍스트 노출 없이 배경을 잘 유지했습니다." }, { "label": "B", "score": 3, "verdict_ko": "인물의 구도와 묘사는 좋으나, 배경 벽면의 마크에 금지된 텍스트가 임의로 생성되는 치명적인 위반이 발생했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 샷의 구도와 인물의 자세를 충실히 구현했으며, 금지된 텍스트 노출 없이 배경을 잘 유지했습니다." }, { "label": "A", "score": 3, "verdict_ko": "인물의 구도와 묘사는 좋으나, 배경 벽면의 마크에 금지된 텍스트가 임의로 생성되는 치명적인 위반이 발생했습니다." } ] }, "combined": { "totals": { "A": 10, "B": 12 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S22sh1__bgfirst_bg.png", "bg_asset_id": "fd3f2282-572e-4d79-83ad-4e4e86c70583", "bg_record_key": "S22sh1::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "경찰서 앞 도로", "groupbg_asset_id": "bb0e0a7c-4bce-45f2-b5b4-94f916494f88" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "groupbg::피시방 좌석": { "input_fingerprint": "abc692c4f7a5d00f", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "피시방 좌석", "tags": [ "S23sh1" ] }, "context_sig": "8b428ef40fe0da73" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: At a computer booth inside a dark internet café, illuminated mainly by the monitor screen.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\nPC방 내부: 모니터 불빛이 주된 조명 역할을 하는 어두운 실내로, 데스크톱 좌석이 밀집된 상업 공간이다. (특징: 일렬로 늘어선 컴퓨터와 모니터; 디지털 지도 화면이 띄워진 모니터; 어두운 조명; 지도 출력물이 인쇄되는 프린터)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#23. 어느 PC방 / 실내\n- 모니터 화면에 앉아있는 수리영\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: At a computer booth inside a dark internet café, illuminated mainly by the monitor screen.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\nPC방 내부: 모니터 불빛이 주된 조명 역할을 하는 어두운 실내로, 데스크톱 좌석이 밀집된 상업 공간이다. (특징: 일렬로 늘어선 컴퓨터와 모니터; 디지털 지도 화면이 띄워진 모니터; 어두운 조명; 지도 출력물이 인쇄되는 프린터)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#23. 어느 PC방 / 실내\n- 모니터 화면에 앉아있는 수리영\n\nTIME OF DAY (lock): day.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_피시방_좌석_f1550f.png", "asset_id": "5b8395c5-4ed7-4b17-8375-5e29768ce9e5", "input_asset_ids": [ "4e718f9c-b2ae-442e-b8e0-982796b470af" ], "origin_tag": "S23sh1", "place_text": "At a computer booth inside a dark internet café, illuminated mainly by the monitor screen.", "origin_inputs": { "place_text": "At a computer booth inside a dark internet café, illuminated mainly by the monitor screen.", "time_of_day_en": "day", "conti_asset_id": "4e718f9c-b2ae-442e-b8e0-982796b470af" } }, "S23sh1::bgfirst_bg": { "input_fingerprint": "624f46d3ce151ee8", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 PC방 안, 모니터 불빛을 받으며 화면을 뚫어지게 응시하는 수리영의 굳은 얼굴.\n\nLOCATION (lock): At a computer booth inside a dark internet café, illuminated mainly by the monitor screen.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From seated eye level just beyond and slightly beside the monitor, the camera completes a slow push toward 수리영 until her rigid face occupies most of the frame. The monitor edge remains soft in the lower foreground while her tense eyes and set jaw hold the upper-center focus, preserving the imminent downward transition toward her hands.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active screen faces 수리영 and away from the camera; only its near edge and partial rear silhouette are visible; used as Soft foreground obstruction separating the camera from 수리영 and motivating the light on her face; PC stations (Dark and inactive-looking beyond the occupied station) — Their monitor edges and seating lines recede behind 수리영; used as Receding background context kept indistinct behind the close facial composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The monitor is the stated key source on 수리영’s face, held in moody low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 어두운 PC방 안, 모니터 불빛을 받으며 화면을 뚫어지게 응시하는 수리영의 굳은 얼굴.\n\nLOCATION (lock): At a computer booth inside a dark internet café, illuminated mainly by the monitor screen.\n\nTIME OF DAY (lock): day.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From seated eye level just beyond and slightly beside the monitor, the camera completes a slow push toward 수리영 until her rigid face occupies most of the frame. The monitor edge remains soft in the lower foreground while her tense eyes and set jaw hold the upper-center focus, preserving the imminent downward transition toward her hands.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active screen faces 수리영 and away from the camera; only its near edge and partial rear silhouette are visible; used as Soft foreground obstruction separating the camera from 수리영 and motivating the light on her face; PC stations (Dark and inactive-looking beyond the occupied station) — Their monitor edges and seating lines recede behind 수리영; used as Receding background context kept indistinct behind the close facial composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The monitor is the stated key source on 수리영’s face, held in moody low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S23sh1__bgfirst_bg.png", "asset_id": "05c5238b-8a7f-40fd-99d6-a44dce32ebdd", "input_asset_ids": [ "4e718f9c-b2ae-442e-b8e0-982796b470af", "5b8395c5-4ed7-4b17-8375-5e29768ce9e5" ] }, "S23sh1": { "input_fingerprint": "443aa4d4139246aa", "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): 어두운 PC방 안, 모니터 불빛을 받으며 화면을 뚫어지게 응시하는 수리영의 굳은 얼굴.\n\nLOCATION (lock): At a computer booth inside a dark internet café, illuminated mainly by the monitor screen. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From seated eye level just beyond and slightly beside the monitor, the camera completes a slow push toward 수리영 until her rigid face occupies most of the frame. The monitor edge remains soft in the lower foreground while her tense eyes and set jaw hold the upper-center focus, preserving the imminent downward transition toward her hands.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active screen faces 수리영 and away from the camera; only its near edge and partial rear silhouette are visible; used as Soft foreground obstruction separating the camera from 수리영 and motivating the light on her face; PC stations (Dark and inactive-looking beyond the occupied station) — Their monitor edges and seating lines recede behind 수리영; used as Receding background context kept indistinct behind the close facial composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The monitor is the stated key source on 수리영’s face, held in moody low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The coordinate note remains available for Su-ri-young to enter the location. She folds the newly printed maps and places them 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두운 PC방 안, 모니터 불빛을 받으며 화면을 뚫어지게 응시하는 수리영의 굳은 얼굴.\n\nLOCATION (lock): At a computer booth inside a dark internet café, illuminated mainly by the monitor screen. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From seated eye level just beyond and slightly beside the monitor, the camera completes a slow push toward 수리영 until her rigid face occupies most of the frame. The monitor edge remains soft in the lower foreground while her tense eyes and set jaw hold the upper-center focus, preserving the imminent downward transition toward her hands.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active screen faces 수리영 and away from the camera; only its near edge and partial rear silhouette are visible; used as Soft foreground obstruction separating the camera from 수리영 and motivating the light on her face; PC stations (Dark and inactive-looking beyond the occupied station) — Their monitor edges and seating lines recede behind 수리영; used as Receding background context kept indistinct behind the close facial composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The monitor is the stated key source on 수리영’s face, held in moody low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The coordinate note remains available for Su-ri-young to enter the location. She folds the newly printed maps and places them 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day.\n\nSHOT TEXT (authoritative, Korean): 어두운 PC방 안, 모니터 불빛을 받으며 화면을 뚫어지게 응시하는 수리영의 굳은 얼굴.\n\nLOCATION (lock): At a computer booth inside a dark internet café, illuminated mainly by the monitor screen. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From seated eye level just beyond and slightly beside the monitor, the camera completes a slow push toward 수리영 until her rigid face occupies most of the frame. The monitor edge remains soft in the lower foreground while her tense eyes and set jaw hold the upper-center focus, preserving the imminent downward transition toward her hands.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: PC monitor (On) — The active screen faces 수리영 and away from the camera; only its near edge and partial rear silhouette are visible; used as Soft foreground obstruction separating the camera from 수리영 and motivating the light on her face; PC stations (Dark and inactive-looking beyond the occupied station) — Their monitor edges and seating lines recede behind 수리영; used as Receding background context kept indistinct behind the close facial composition.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The monitor is the stated key source on 수리영’s face, held in moody low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The coordinate note remains available for Su-ri-young to enter the location. She folds the newly printed maps and places them 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S23sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S23sh1.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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_피시방_좌석_f1550f.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" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "피사체 약간 측면의 카메라 위치와 전경의 부드러운 모니터 아웃포커싱 등 프레이밍 지시를 가장 잘 구현했으며, 위치 디테일과 소품(인형 달린 백팩) 묘사도 훌륭합니다 (단, 의상 색상 불일치)." }, { "label": "A", "score": 5, "verdict_ko": "캐릭터 의상은 일치하나, 완전한 정면 구도와 하단부의 선명하고 날카로운 검은색 실루엣(모니터)을 렌더링하여 '약간 측면' 및 '부드러운 전경 장애물'이라는 높은 우선순위의 카메라/프레이밍 지시를 어겼습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "피사체 약간 측면의 카메라 위치와 전경의 부드러운 모니터 아웃포커싱 등 프레이밍 지시를 가장 잘 구현했으며, 위치 디테일과 소품(인형 달린 백팩) 묘사도 훌륭합니다 (단, 의상 색상 불일치)." }, { "label": "A", "score": 5, "verdict_ko": "캐릭터 의상은 일치하나, 완전한 정면 구도와 하단부의 선명하고 날카로운 검은색 실루엣(모니터)을 렌더링하여 '약간 측면' 및 '부드러운 전경 장애물'이라는 높은 우선순위의 카메라/프레이밍 지시를 어겼습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "부드러운 모니터 전경을 활용한 클로즈업 프레이밍, 캐릭터의 레퍼런스 의상(회색 후드티), 그리고 모니터 불빛이 얼굴에 떨어지는 로우키 조명 분위기를 아주 정확하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "프레이밍과 소품(인형 달린 가방)은 요구사항을 반영했으나, 캐릭터 레퍼런스의 의상(회색 후드티)과 일치하지 않으며 텍스트가 요구한 모니터 주광원 조명 연출이 부족합니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "부드러운 모니터 전경을 활용한 클로즈업 프레이밍, 캐릭터의 레퍼런스 의상(회색 후드티), 그리고 모니터 불빛이 얼굴에 떨어지는 로우키 조명 분위기를 아주 정확하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "프레이밍과 소품(인형 달린 가방)은 요구사항을 반영했으나, 캐릭터 레퍼런스의 의상(회색 후드티)과 일치하지 않으며 텍스트가 요구한 모니터 주광원 조명 연출이 부족합니다." } ] }, "combined": { "totals": { "A": 14, "B": 12 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 14, "B": 12 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "피사체 약간 측면의 카메라 위치와 전경의 부드러운 모니터 아웃포커싱 등 프레이밍 지시를 가장 잘 구현했으며, 위치 디테일과 소품(인형 달린 백팩) 묘사도 훌륭합니다 (단, 의상 색상 불일치)." }, { "label": "A", "score": 5, "verdict_ko": "캐릭터 의상은 일치하나, 완전한 정면 구도와 하단부의 선명하고 날카로운 검은색 실루엣(모니터)을 렌더링하여 '약간 측면' 및 '부드러운 전경 장애물'이라는 높은 우선순위의 카메라/프레이밍 지시를 어겼습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_피시방_좌석_f1550f.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물이 레이아웃 스케치의 지정된 위치(화면 우측)에 있지 않고 중앙에 배치되어 있으며, 시선 방향 역시 스케치의 좌측 하단이 아닌 정면 아래를 향하고 있습니다.", "fix_en": "Character is incorrectly centered instead of positioned on the right, and her gaze points straight down rather than down-left per the sketch.", "unfixable": true }, { "issue_ko": "원본 배경 이미지를 그대로 유지해야 한다는 지침을 위반하여, 우측에 있던 프린터가 사라지고 헤드폰이 걸린 파티션으로 대체되는 등 배경 구조가 완전히 변형되었습니다.", "fix_en": "The original background reference was not kept exactly; the right-side printer is missing and the room's desk geometry has been completely altered.", "unfixable": true } ] }, "composition_critique": { "issues": [ { "issue_ko": "프레이밍이 요구된 클로즈업보다 넓어 수리영의 굳은 얼굴이 화면 대부분을 차지하지 못하고, 전경 모니터가 하단의 지나치게 큰 영역을 가린다.", "fix_en": "Enlarge Su-ri-young within the existing frame so her face dominates the upper center, while minimizing the visible foreground monitor area without changing the viewpoint or background geometry.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Enlarge Su-ri-young within the existing frame so her face dominates the upper center, while minimizing the visible foreground monitor area without changing the viewpoint or background geometry.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 대로 원본 배경 이미지의 디테일을 완벽하게 유지하면서 캐릭터의 인상과 모니터 조명을 사실적으로 합성했습니다." }, { "label": "B", "score": 4, "verdict_ko": "캐릭터의 구도와 표정은 스케치와 잘 맞으나, 배경을 정확히 유지하라는 지시를 위반하여 우측 프린터 등의 디테일이 훼손되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 대로 원본 배경 이미지의 디테일을 완벽하게 유지하면서 캐릭터의 인상과 모니터 조명을 사실적으로 합성했습니다." }, { "label": "B", "score": 4, "verdict_ko": "캐릭터의 구도와 표정은 스케치와 잘 맞으나, 배경을 정확히 유지하라는 지시를 위반하여 우측 프린터 등의 디테일이 훼손되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "레이아웃 스케치의 클로즈업 구도와 인물 크기를 정확히 반영하였으며, 지시된 조명과 분위기를 충실히 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "인물의 크기가 작고 모니터가 프레임을 과도하게 가려, 얼굴이 화면 대부분을 차지해야 한다는 스케치 및 프레이밍 지침을 어겼습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "레이아웃 스케치의 클로즈업 구도와 인물 크기를 정확히 반영하였으며, 지시된 조명과 분위기를 충실히 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "인물의 크기가 작고 모니터가 프레임을 과도하게 가려, 얼굴이 화면 대부분을 차지해야 한다는 스케치 및 프레이밍 지침을 어겼습니다." } ] }, "combined": { "totals": { "A": 12, "B": 12 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S23sh1__bgfirst_bg.png", "bg_asset_id": "05c5238b-8a7f-40fd-99d6-a44dce32ebdd", "bg_record_key": "S23sh1::bgfirst_bg", "chain_winner": true, "authority": "groupbg", "group_key": "피시방 좌석", "groupbg_asset_id": "5b8395c5-4ed7-4b17-8375-5e29768ce9e5" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" } }, "S24sh2::bgfirst_bg": { "input_fingerprint": "f36c2fa75026620d", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 비에 흠뻑 젖은 채 앞을 직시하는 수리영의 결연한 눈빛 클로즈업.\n\nLOCATION (lock): An exposed coastal roadway in a heavy thunderstorm, with driving rain and dark clouds overhead.\n\nTIME OF DAY (lock): day, stormy.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level directly ahead of the bicycle, the camera tracks backward at matching speed, holding her rain-soaked face tightly while the moving surroundings register only at the frame edges. Her forward gaze runs just past the lens, making the near-frontal angle read as determination rather than direct address.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Coastal route (Under heavy rain); used as Peripheral motion context behind and beside the tightly framed rider; Heavy rainfall (Falling forcefully); used as Visible streaks and impacts across 수리영 and the surrounding frame; Clouded sky (Heavily overcast with thunder and lightning); used as Storm background establishing the severe weather around the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-dark ambient light and the explicitly present lightning create hard intermittent contrast across the rain-soaked face, with otherwise muted neutral color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 비에 흠뻑 젖은 채 앞을 직시하는 수리영의 결연한 눈빛 클로즈업.\n\nLOCATION (lock): An exposed coastal roadway in a heavy thunderstorm, with driving rain and dark clouds overhead.\n\nTIME OF DAY (lock): day, stormy.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level directly ahead of the bicycle, the camera tracks backward at matching speed, holding her rain-soaked face tightly while the moving surroundings register only at the frame edges. Her forward gaze runs just past the lens, making the near-frontal angle read as determination rather than direct address.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Coastal route (Under heavy rain); used as Peripheral motion context behind and beside the tightly framed rider; Heavy rainfall (Falling forcefully); used as Visible streaks and impacts across 수리영 and the surrounding frame; Clouded sky (Heavily overcast with thunder and lightning); used as Storm background establishing the severe weather around the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-dark ambient light and the explicitly present lightning create hard intermittent contrast across the rain-soaked face, with otherwise muted neutral color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S24sh2__bgfirst_bg.png", "asset_id": "7b90ace8-177c-40dc-9fcc-bea6c3d09fb0", "input_asset_ids": [ "dd017c5e-bed4-4fae-82ef-1cbeb6502602", "8f0394d6-f50d-4f43-baa3-d2420dc9b871" ] }, "S24sh2": { "input_fingerprint": "e100e214dc39a708", "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, stormy.\n\nSHOT TEXT (authoritative, Korean): 비에 흠뻑 젖은 채 앞을 직시하는 수리영의 결연한 눈빛 클로즈업.\n\nLOCATION (lock): An exposed coastal roadway in a heavy thunderstorm, with driving rain and dark clouds overhead. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level directly ahead of the bicycle, the camera tracks backward at matching speed, holding her rain-soaked face tightly while the moving surroundings register only at the frame edges. Her forward gaze runs just past the lens, making the near-frontal angle read as determination rather than direct address.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Coastal route (Under heavy rain); used as Peripheral motion context behind and beside the tightly framed rider; Heavy rainfall (Falling forcefully); used as Visible streaks and impacts across 수리영 and the surrounding frame; Clouded sky (Heavily overcast with thunder and lightning); used as Storm background establishing the severe weather around the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-dark ambient light and the explicitly present lightning create hard intermittent contrast across the rain-soaked face, with otherwise muted neutral color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young continues on her bicycle with the printed maps secured inside her backpack, which remains with her in the rain.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, stormy.\n\nSHOT TEXT (authoritative, Korean): 비에 흠뻑 젖은 채 앞을 직시하는 수리영의 결연한 눈빛 클로즈업.\n\nLOCATION (lock): An exposed coastal roadway in a heavy thunderstorm, with driving rain and dark clouds overhead. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level directly ahead of the bicycle, the camera tracks backward at matching speed, holding her rain-soaked face tightly while the moving surroundings register only at the frame edges. Her forward gaze runs just past the lens, making the near-frontal angle read as determination rather than direct address.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Coastal route (Under heavy rain); used as Peripheral motion context behind and beside the tightly framed rider; Heavy rainfall (Falling forcefully); used as Visible streaks and impacts across 수리영 and the surrounding frame; Clouded sky (Heavily overcast with thunder and lightning); used as Storm background establishing the severe weather around the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-dark ambient light and the explicitly present lightning create hard intermittent contrast across the rain-soaked face, with otherwise muted neutral color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young continues on her bicycle with the printed maps secured inside her backpack, which remains with her in the rain.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): day, stormy.\n\nSHOT TEXT (authoritative, Korean): 비에 흠뻑 젖은 채 앞을 직시하는 수리영의 결연한 눈빛 클로즈업.\n\nLOCATION (lock): An exposed coastal roadway in a heavy thunderstorm, with driving rain and dark clouds overhead. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At 수리영’s eye level directly ahead of the bicycle, the camera tracks backward at matching speed, holding her rain-soaked face tightly while the moving surroundings register only at the frame edges. Her forward gaze runs just past the lens, making the near-frontal angle read as determination rather than direct address.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Coastal route (Under heavy rain); used as Peripheral motion context behind and beside the tightly framed rider; Heavy rainfall (Falling forcefully); used as Visible streaks and impacts across 수리영 and the surrounding frame; Clouded sky (Heavily overcast with thunder and lightning); used as Storm background establishing the severe weather around the rider.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Storm-dark ambient light and the explicitly present lightning create hard intermittent contrast across the rain-soaked face, with otherwise muted neutral color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young continues on her bicycle with the printed maps secured inside her backpack, which remains with her in the rain.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S24sh2__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S24sh2.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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해안_도로_8394b3.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" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "얼굴을 타이트하게 잡는 정확한 클로즈업 샷을 구현하여 프레이밍 우선순위를 완벽히 준수했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지시된 클로즈업 샷을 무시하고 배낭과 상체를 보여주기 위해 프레임을 넓힌 점이 감점 요인입니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "얼굴을 타이트하게 잡는 정확한 클로즈업 샷을 구현하여 프레이밍 우선순위를 완벽히 준수했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지시된 클로즈업 샷을 무시하고 배낭과 상체를 보여주기 위해 프레임을 넓힌 점이 감점 요인입니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "명시된 '클로즈업' 샷 크기를 완벽히 준수하여 빗물에 젖은 얼굴을 타이트하게 포착했으며, 올바른 프레이밍으로 인해 배낭이 보이지 않는 점은 감점되지 않습니다." }, { "label": "B", "score": 4, "verdict_ko": "지정된 타이트한 클로즈업을 무시하고 배낭과 지도를 보여주기 위해 프레임을 넓혀 미디엄 샷으로 연출한 것은 프레이밍 지침 위반입니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "명시된 '클로즈업' 샷 크기를 완벽히 준수하여 빗물에 젖은 얼굴을 타이트하게 포착했으며, 올바른 프레이밍으로 인해 배낭이 보이지 않는 점은 감점되지 않습니다." }, { "label": "A", "score": 4, "verdict_ko": "지정된 타이트한 클로즈업을 무시하고 배낭과 지도를 보여주기 위해 프레임을 넓혀 미디엄 샷으로 연출한 것은 프레이밍 지침 위반입니다." } ] }, "combined": { "totals": { "A": 9, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 9, "B": 15 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "얼굴을 타이트하게 잡는 정확한 클로즈업 샷을 구현하여 프레이밍 우선순위를 완벽히 준수했습니다." }, { "label": "A", "score": 5, "verdict_ko": "지시된 클로즈업 샷을 무시하고 배낭과 상체를 보여주기 위해 프레임을 넓힌 점이 감점 요인입니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_해안_도로_8394b3.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "시선이 렌즈를 살짝 비껴가야 한다는 지시와 달리 카메라를 정면으로 응시하고 있습니다.", "fix_en": "Shift her pupils slightly to the left or right so her gaze runs just past the lens instead of making direct eye contact.", "unfixable": false }, { "issue_ko": "번개로 인해 젖은 얼굴에 강한 명암 대비가 생겨야 하지만, 얼굴의 조명이 너무 부드럽고 평면적입니다.", "fix_en": "Add a hard, bright highlight to one side of her face to simulate the intermittent, high-contrast flash of lightning.", "unfixable": false }, { "issue_ko": "배낭을 메고 이동 중이라는 설정임에도 프레임 안으로 보이는 어깨에 배낭 끈이 없습니다.", "fix_en": "Add wet, dark backpack straps running down over the front of both her shoulders.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Shift her gaze subtly just past the lens while preserving the near-frontal head angle and all other composition.", "issue_ko": "시선이 렌즈를 직접 응시해 관객에게 말을 거는 구도로 보이며, 계약에서 지정한 렌즈 바로 옆을 향하는 전방 시선과 다릅니다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Shift her pupils slightly to the left or right so her gaze runs just past the lens instead of making direct eye contact.\n- Add a hard, bright highlight to one side of her face to simulate the intermittent, high-contrast flash of lightning.\n- Add wet, dark backpack straps running down over the front of both her shoulders.\n- Shift her gaze subtly just past the lens while preserving the near-frontal head angle and all other composition.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "렌즈를 살짝 빗겨가는 시선 처리로 결연함을 표현하라는 카메라 지시를 정확히 따랐으며, 배낭의 디테일도 잘 살렸습니다." }, { "label": "A", "score": 5, "verdict_ko": "비에 젖은 클로즈업 묘사와 배경은 훌륭하나, 프롬프트에서 피하라고 명시한 렌즈와의 직접적인 시선 교환(direct address)을 연출하여 감점되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "렌즈를 살짝 빗겨가는 시선 처리로 결연함을 표현하라는 카메라 지시를 정확히 따랐으며, 배낭의 디테일도 잘 살렸습니다." }, { "label": "A", "score": 5, "verdict_ko": "비에 젖은 클로즈업 묘사와 배경은 훌륭하나, 프롬프트에서 피하라고 명시한 렌즈와의 직접적인 시선 교환(direct address)을 연출하여 감점되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 8, "verdict_ko": "렌즈를 살짝 비껴가는 시선 처리와 배낭 끈 묘사로 카메라 지시사항과 착장 상태를 정확히 충족했습니다." }, { "label": "B", "score": 5, "verdict_ko": "렌즈를 정면으로 응시하여 '직접적인 시선 교환(direct address)을 피하라'는 구체적인 지시를 위반했으며 배낭 끈도 보이지 않습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "렌즈를 살짝 비껴가는 시선 처리와 배낭 끈 묘사로 카메라 지시사항과 착장 상태를 정확히 충족했습니다." }, { "label": "A", "score": 5, "verdict_ko": "렌즈를 정면으로 응시하여 '직접적인 시선 교환(direct address)을 피하라'는 구체적인 지시를 위반했으며 배낭 끈도 보이지 않습니다." } ] }, "combined": { "totals": { "A": 10, "B": 16 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S24sh2__bgfirst_bg.png", "bg_asset_id": "7b90ace8-177c-40dc-9fcc-bea6c3d09fb0", "bg_record_key": "S24sh2::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "해안 도로", "groupbg_asset_id": "8f0394d6-f50d-4f43-baa3-d2420dc9b871" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S25sh1::variants": { "author_fp": "7b780c30825db3d2", "author": { "variants": [ { "approach_ko": "바닥과 수평을 맞춘 깊은 초점의 광각으로 발자국의 선을 따라 난장판 전체와 열린 창까지 냉정하게 읽히게 한다.", "prompt_en": "Use a floor-skimming, level wide-angle perspective with strong rectilinear depth, keeping the footprint trail crisp as it pulls the eye through the overturned chairs to the open window. Hold generous depth of field so the room’s disturbed geometry remains fully legible, with subdued dusk light falling off naturally into muted, low-key interior shadow." }, { "approach_ko": "불안하게 기운 저각도와 겹치는 의자 다리로 침입의 여파를 압박감 있게 층층이 드러낸다.", "prompt_en": "Give the low wide frame a restrained off-level tilt, using the irregular chair seats and legs as sharp midground diagonals that interrupt but never conceal the footprint trail toward the open window. Build pronounced near-to-far layering and let the dim dusk opening provide a soft background lift against the heavier interior, creating a tense, unsettled sense of intrusion while preserving clear spatial readability." } ] }, "reused": false }, "S25sh1": { "input_fingerprint": "25dd008a1590d382", "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): 해 질 무렵의 옥탑방 거실, 바닥에 찍힌 흙 묻은 발자국과 쓰러진 의자들이 널브러진 난장판 구도.\n\nLOCATION (lock): The ransacked living and dining area inside the rooftop home at dusk, with an open window, muddy footprints, and overturned chairs. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just above the living-room floor near a corner, a low wide composition begins to track along the muddy footprints toward the open window and overturned dining chairs. The footprint trail crosses the lower frame as the room’s disordered layout remains fully legible, preparing the later rise toward 수리영’s room.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Muddy footprints (Crossing the living-room floor); used as Primary leading line carrying the camera through the disordered room; Dining chairs (Overturned) — Their seats and legs are exposed at irregular sideways angles toward the camera; used as Midground evidence of the search and intrusion; Living-room window (Open) — The window is open, with the exterior beyond visible through the opening; used as Background endpoint of the initial low tracking line; Household furnishings (Ransacked) — Different displaced sides and openings face the low camera; used as Environmental evidence surrounding the footprint path.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience renders the ransacked interior in low-key, muted-neutral contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room has been ransacked: muddy footprints remain on the floor, the living-room window is open, and the dining chairs are overturned. The red circle and bloodied photograph are already fixed to the bedroom wall before Su-ri-young discovers them.\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): 해 질 무렵의 옥탑방 거실, 바닥에 찍힌 흙 묻은 발자국과 쓰러진 의자들이 널브러진 난장판 구도.\n\nLOCATION (lock): The ransacked living and dining area inside the rooftop home at dusk, with an open window, muddy footprints, and overturned chairs. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just above the living-room floor near a corner, a low wide composition begins to track along the muddy footprints toward the open window and overturned dining chairs. The footprint trail crosses the lower frame as the room’s disordered layout remains fully legible, preparing the later rise toward 수리영’s room.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Muddy footprints (Crossing the living-room floor); used as Primary leading line carrying the camera through the disordered room; Dining chairs (Overturned) — Their seats and legs are exposed at irregular sideways angles toward the camera; used as Midground evidence of the search and intrusion; Living-room window (Open) — The window is open, with the exterior beyond visible through the opening; used as Background endpoint of the initial low tracking line; Household furnishings (Ransacked) — Different displaced sides and openings face the low camera; used as Environmental evidence surrounding the footprint path.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience renders the ransacked interior in low-key, muted-neutral contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room has been ransacked: muddy footprints remain on the floor, the living-room window is open, and the dining chairs are overturned. The red circle and bloodied photograph are already fixed to the bedroom wall before Su-ri-young discovers them.\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 floor-skimming, level wide-angle perspective with strong rectilinear depth, keeping the footprint trail crisp as it pulls the eye through the overturned chairs to the open window. Hold generous depth of field so the room’s disturbed geometry remains fully legible, with subdued dusk light falling off naturally into muted, low-key interior 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): dusk.\n\nSHOT TEXT (authoritative, Korean): 해 질 무렵의 옥탑방 거실, 바닥에 찍힌 흙 묻은 발자국과 쓰러진 의자들이 널브러진 난장판 구도.\n\nLOCATION (lock): The ransacked living and dining area inside the rooftop home at dusk, with an open window, muddy footprints, and overturned chairs. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From just above the living-room floor near a corner, a low wide composition begins to track along the muddy footprints toward the open window and overturned dining chairs. The footprint trail crosses the lower frame as the room’s disordered layout remains fully legible, preparing the later rise toward 수리영’s room.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Muddy footprints (Crossing the living-room floor); used as Primary leading line carrying the camera through the disordered room; Dining chairs (Overturned) — Their seats and legs are exposed at irregular sideways angles toward the camera; used as Midground evidence of the search and intrusion; Living-room window (Open) — The window is open, with the exterior beyond visible through the opening; used as Background endpoint of the initial low tracking line; Household furnishings (Ransacked) — Different displaced sides and openings face the low camera; used as Environmental evidence surrounding the footprint path.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience renders the ransacked interior in low-key, muted-neutral contrast.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room has been ransacked: muddy footprints remain on the floor, the living-room window is open, and the dining chairs are overturned. The red circle and bloodied photograph are already fixed to the bedroom wall before Su-ri-young discovers them.\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):\nGive the low wide frame a restrained off-level tilt, using the irregular chair seats and legs as sharp midground diagonals that interrupt but never conceal the footprint trail toward the open window. Build pronounced near-to-far layering and let the dim dusk opening provide a soft background lift against the heavier interior, creating a tense, unsettled sense of intrusion while preserving clear spatial readability.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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 로우 앵글 구도와 주요 요소(진흙 발자국, 열린 창문, 넘어진 의자)를 잘 배치하여 프롬프트를 충실히 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "요청하지 않은 현관문이 열려 있으며, 우측에 위치한 넘어진 의자의 기하학적 구조가 물리적으로 불가능하게 왜곡되어 치명적인 오류가 발생했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B01.png" } ], "critique": { "issues": [ { "issue_ko": "열린 창문 너머로 밝은 대낮의 하늘이 보여, 프롬프트가 지정한 '해 질 무렵(dusk)' 시간대와 일치하지 않습니다.", "fix_en": "Change the bright daytime sky visible through the open window to a darker, subdued dusk sky.", "unfixable": false }, { "issue_ko": "바닥에 넘어진 식탁 의자들의 금속 프레임이 쿠션을 뚫고 지나가는 등 물리적으로 불가능한 기하학적 오류가 있습니다.", "fix_en": "Redraw the overturned dining chairs with physically correct geometry, ensuring the metal legs and frames attach naturally to the undersides of the seats without passing through solid materials.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Change the bright daytime sky visible through the open window to a darker, subdued dusk sky.\n- Redraw the overturned dining chairs with physically correct geometry, ensuring the metal legs and frames attach naturally to the undersides of the seats without passing through solid materials.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 로우 앵글과 난장판 구도는 잘 구현했으나, 채광이 너무 밝아 '해 질 무렵(dusk)'과 '어두운 실내(low-key)'라는 조명 지시를 충족하지 못했습니다." }, { "label": "B", "score": 10, "verdict_ko": "창밖의 해 질 무렵 하늘과 그에 따른 어두운 실내 조명(low-key)을 정확히 연출했으며, 엎어진 의자와 진흙 발자국 등 지시된 모든 요소를 완벽하게 반영했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 로우 앵글과 난장판 구도는 잘 구현했으나, 채광이 너무 밝아 '해 질 무렵(dusk)'과 '어두운 실내(low-key)'라는 조명 지시를 충족하지 못했습니다." }, { "label": "B", "score": 10, "verdict_ko": "창밖의 해 질 무렵 하늘과 그에 따른 어두운 실내 조명(low-key)을 정확히 연출했으며, 엎어진 의자와 진흙 발자국 등 지시된 모든 요소를 완벽하게 반영했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 10, "verdict_ko": "지정된 로우 앵글 구도와 난장판이 된 실내 묘사를 훌륭하게 구현했으며, 창밖으로 해 질 무렵(dusk)의 하늘을 정확히 표현했습니다." }, { "label": "B", "score": 7, "verdict_ko": "구도와 방 안의 상태는 잘 묘사되었으나, 창밖의 하늘이 너무 밝아 필수 조건인 '해 질 무렵(dusk)'을 제대로 반영하지 못했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "지정된 로우 앵글 구도와 난장판이 된 실내 묘사를 훌륭하게 구현했으며, 창밖으로 해 질 무렵(dusk)의 하늘을 정확히 표현했습니다." }, { "label": "A", "score": 7, "verdict_ko": "구도와 방 안의 상태는 잘 묘사되었으나, 창밖의 하늘이 너무 밝아 필수 조건인 '해 질 무렵(dusk)'을 제대로 반영하지 못했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 20 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B03", "choice": "Candidate C", "confident": true, "reason_ko": "원문에서 '옥탑방 거실'과 '바닥에 찍힌 흙 묻은 발자국'을 명시하고 있으며, 해당 서브공간(거실 및 주방 구역)을 보여주는 기존 할당 컷이 지시된 공간과 일치하므로 이를 유지합니다.", "kept": "L04B03" } }, "S25sh3::variants": { "author_fp": "f39042d06619cc85", "author": { "variants": [ { "approach_ko": "정면 대칭과 깊은 초점으로 방문·붉은 원·사진을 하나의 차가운 증거 구조로 포착한다.", "prompt_en": "Use a restrained, dead-on architectural composition with the doorway edges forming a balanced outer bracket around the wall. Keep the photograph, red circle, wall texture, and inner doorway surfaces clearly resolved in deep focus, creating a clinical evidentiary stillness. Let the subdued dusk illumination fall softly across the wall, preserving muted neutral tones while the red mark and blood remain the only strong color accents." }, { "approach_ko": "얕은 초점과 압축된 원근으로 흐릿한 방문 사이의 피 묻은 사진을 불길하게 고립시킨다.", "prompt_en": "Photograph the same axial close-up with a compressed, claustrophobic lens feel and very selective focus locked on the bloodstained photograph and the red strokes immediately around it. Allow the doorway edges in the near foreground and the surrounding wall plane to soften gradually, visually tunneling attention toward the evidence while retaining their framing presence. Use faint raking dusk light to reveal the photograph’s worn surface, pinning, blood gloss, and the physical stroke texture of the circle against deep, subdued falloff." } ] }, "reused": false }, "S25sh3": { "input_fingerprint": "a61153439b4539b4", "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): 수리영의 방문 너머, 벽면에 그려진 붉은 원 한가운데 피 묻은 사진이 꽂혀 있는 모습.\n\nLOCATION (lock): Inside the young woman's bedroom, focused on the wall bearing a red circle and a bloodstained photograph in the fading dusk light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level within the bedroom doorway, the camera advances straight toward the wall while narrowing from the room context to the red circle and its pinned photograph. The photograph remains below forty percent of the frame, centered within the larger red mark so that the wall and doorway still anchor its physical scale.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Bloodstained old photograph (Bloodstained and pinned to the wall) — Its image-bearing front faces the camera, showing an old rural restaurant, two seated girls, and the small written detail associated with 금월; used as Central focal evidence framed by the surrounding wall mark; Red circle on the wall (Drawn around the pinned photograph); used as Graphic frame surrounding and isolating the photograph; Open bedroom doorway (Open) — The doorway’s inner edges face the camera and bracket the target wall; used as Outer architectural frame that channels the push toward the wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambience remains subdued and low-key, with the existing red circle and blood providing the only pronounced color accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The ransacked condition, muddy footprints, open window, and fallen chairs remain. A bloodied old photograph is pinned at the center of the red circle on Su-ri-young’s bedroom wall.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 방문 너머, 벽면에 그려진 붉은 원 한가운데 피 묻은 사진이 꽂혀 있는 모습.\n\nLOCATION (lock): Inside the young woman's bedroom, focused on the wall bearing a red circle and a bloodstained photograph in the fading dusk light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level within the bedroom doorway, the camera advances straight toward the wall while narrowing from the room context to the red circle and its pinned photograph. The photograph remains below forty percent of the frame, centered within the larger red mark so that the wall and doorway still anchor its physical scale.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Bloodstained old photograph (Bloodstained and pinned to the wall) — Its image-bearing front faces the camera, showing an old rural restaurant, two seated girls, and the small written detail associated with 금월; used as Central focal evidence framed by the surrounding wall mark; Red circle on the wall (Drawn around the pinned photograph); used as Graphic frame surrounding and isolating the photograph; Open bedroom doorway (Open) — The doorway’s inner edges face the camera and bracket the target wall; used as Outer architectural frame that channels the push toward the wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambience remains subdued and low-key, with the existing red circle and blood providing the only pronounced color accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The ransacked condition, muddy footprints, open window, and fallen chairs remain. A bloodied old photograph is pinned at the center of the red circle on Su-ri-young’s bedroom wall.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, dead-on architectural composition with the doorway edges forming a balanced outer bracket around the wall. Keep the photograph, red circle, wall texture, and inner doorway surfaces clearly resolved in deep focus, creating a clinical evidentiary stillness. Let the subdued dusk illumination fall softly across the wall, preserving muted neutral tones while the red mark and blood remain the only strong color accents.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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): 수리영의 방문 너머, 벽면에 그려진 붉은 원 한가운데 피 묻은 사진이 꽂혀 있는 모습.\n\nLOCATION (lock): Inside the young woman's bedroom, focused on the wall bearing a red circle and a bloodstained photograph in the fading dusk light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At eye level within the bedroom doorway, the camera advances straight toward the wall while narrowing from the room context to the red circle and its pinned photograph. The photograph remains below forty percent of the frame, centered within the larger red mark so that the wall and doorway still anchor its physical scale.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Bloodstained old photograph (Bloodstained and pinned to the wall) — Its image-bearing front faces the camera, showing an old rural restaurant, two seated girls, and the small written detail associated with 금월; used as Central focal evidence framed by the surrounding wall mark; Red circle on the wall (Drawn around the pinned photograph); used as Graphic frame surrounding and isolating the photograph; Open bedroom doorway (Open) — The doorway’s inner edges face the camera and bracket the target wall; used as Outer architectural frame that channels the push toward the wall.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dusk ambience remains subdued and low-key, with the existing red circle and blood providing the only pronounced color accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The ransacked condition, muddy footprints, open window, and fallen chairs remain. A bloodied old photograph is pinned at the center of the red circle on Su-ri-young’s bedroom wall.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the same axial close-up with a compressed, claustrophobic lens feel and very selective focus locked on the bloodstained photograph and the red strokes immediately around it. Allow the doorway edges in the near foreground and the surrounding wall plane to soften gradually, visually tunneling attention toward the evidence while retaining their framing presence. Use faint raking dusk light to reveal the photograph’s worn surface, pinning, blood gloss, and the physical stroke texture of the circle against deep, subdued falloff.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "A": 2, "B": 9 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "방의 구조가 양옆으로 복제되는 물리적으로 불가능한 건축 형태를 보여주어 심각한 하드 위반입니다." }, { "label": "B", "score": 9, "verdict_ko": "문틈 사이로 붉은 원과 사진에 집중하는 클로즈업 구도를 정확히 구현했으며, 요구된 공간의 분위기와 디테일을 훌륭하게 반영했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L04B02.png" } ], "critique": { "issues": [ { "issue_ko": "붉은 원이 사진 뒤쪽까지 완전히 칠해진 형태입니다. 프롬프트의 지시대로 속이 빈 고리 모양의 선을 유지해야 합니다.", "fix_en": "Remove the solid red paint behind and immediately around the photograph, replacing it with the bare wall texture so the red mark forms an open ring of brush-width.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the solid red paint behind and immediately around the photograph, replacing it with the bare wall texture so the red mark forms an open ring of brush-width.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "문틈 사이로 보이는 앵글과 속이 빈 고리 형태의 붉은 원, 그리고 중앙에 핀으로 고정된 피 묻은 사진을 지침에 맞게 완벽하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "벽에 그려진 붉은 원이 속이 꽉 찬 형태로 묘사되어, 내부 벽면이 보여야 하는 붓 자국 형태의 빈 고리로 그리라는 명시적 지침을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "문틈 사이로 보이는 앵글과 속이 빈 고리 형태의 붉은 원, 그리고 중앙에 핀으로 고정된 피 묻은 사진을 지침에 맞게 완벽하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "벽에 그려진 붉은 원이 속이 꽉 찬 형태로 묘사되어, 내부 벽면이 보여야 하는 붓 자국 형태의 빈 고리로 그리라는 명시적 지침을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "붉은 원을 속이 빈 고리 형태로 그리라는 명시적 규칙을 정확히 준수했으며, 문틀을 활용한 클로즈업 구도와 피 묻은 사진의 디테일을 잘 구현했습니다." }, { "label": "B", "score": 5, "verdict_ko": "요구된 샷과 프레이밍은 잘 구현했으나, 붉은 원의 내부가 완전히 채워져 있어 '채워졌다는 명시가 없는 한 빈 고리 형태를 유지하라'는 지침을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "붉은 원을 속이 빈 고리 형태로 그리라는 명시적 규칙을 정확히 준수했으며, 문틀을 활용한 클로즈업 구도와 피 묻은 사진의 디테일을 잘 구현했습니다." }, { "label": "A", "score": 5, "verdict_ko": "요구된 샷과 프레이밍은 잘 구현했으나, 붉은 원의 내부가 완전히 채워져 있어 '채워졌다는 명시가 없는 한 빈 고리 형태를 유지하라'는 지침을 위반했습니다." } ] }, "combined": { "totals": { "A": 9, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L04B01", "B": "L04B02", "C": "L04B03", "D": "L04B04", "E": "L04B05", "F": "L04B06", "G": "L04B07" }, "assigned": "L04B02", "choice": "Candidate B", "confident": true, "reason_ko": "원문에서 '방문 너머, 벽면에 그려진 붉은 원 한가운데 피 묻은 사진'을 묘사하고 있으므로, 침실 벽면에 붉은 원과 사진이 있는 현재 할당된 이미지가 정확히 일치합니다.", "kept": "L04B02" } }, "S25sh7::variants": { "author_fp": "8c4f83596f395cf7", "author": { "variants": [ { "approach_ko": "깊은 초점과 정면의 출구 프레임으로 떠나는 인물과 뒤에 남은 난장판을 동시에 선명하게 포착한다.", "prompt_en": "Use a restrained, straight-ahead wide composition with a natural wide-lens feel, holding her backpack and hunched shoulders firmly at center as the doorway forms a severe frame around her departure. Keep generous depth of field so the crumpled photograph at her side, the entrance edges, and the irregularly angled disorder receding behind her remain materially legible. Let the subdued dusk from ahead and the existing window ambience separate her darkened back from the exit without turning her into a featureless silhouette." }, { "approach_ko": "얕은 초점과 압축된 깊이감으로 움켜쥔 사진과 무거운 뒷모습에 감정적 시선을 집중한다.", "prompt_en": "Photograph the wide shot with a gently compressed lens feel and selective focus concentrated across her upper back, backpack, clenched hand, and the partly exposed face of the crumpled photograph. Allow the ransacked interior to fall away in softer, overlapping layers while its overturned and displaced forms remain unmistakable around the route to the open door. Shape the low dusk ambience with delicate edge separation on fabric, fingers, and wrinkled paper, emphasizing the weight and finality of the step outward." } ] }, "reused": false }, "S25sh7": { "input_fingerprint": "ed446c533cc0ea05", "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): 백팩을 멘 채 한 손에 구겨진 사진을 쥐고 열린 현관 밖으로 발을 내디딘 수리영의 뒷모습.\n\nLOCATION (lock): At the rooftop home's interior entryway, crossing through the open front door from the ransacked home at dusk. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-back height one step behind 수리영, the camera completes its following move through the living room as she crosses the open entrance. Her backpack and hunched shoulders occupy the central midground, while the hand gripping the crumpled photograph hangs clearly along one side and the disturbed room recedes behind her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Entrance door (Open) — Its open inner face and doorway edges lead away from the camera; used as Exit frame directly ahead of 수리영 and endpoint of the tracking path; Ransacked living-room furnishings (Disordered) — Displaced sides and overturned forms face the following camera at irregular angles; used as Background evidence left behind as 수리영 exits; Old photograph (Crumpled, bloodstained, and removed from the wall) — The crumpled image-bearing face is only partly visible between her clenched fingers; used as Small but readable object held beside 수리영’s body; Backpack (Worn on 수리영’s back) — Its rear face is presented directly toward the following camera; used as Silhouette and load defining 수리영’s departure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience holds the interior in muted low-key contrast as 수리영 moves through the open entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 sunset interior light, open window, muddy footprints, overturned chairs, and ransacked living-room state from the reference. Preserve the young woman's clothing, backpack, and crumpled photograph, but exclude the bedroom's red circle and wall-mounted photograph from the exit 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bloodied photograph has been removed from the wall and is tightly held in Su-ri-young’s hand, while the red wall circle and the ransacked room remain behind. She leaves wearing the backpack that contains her printed maps.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 백팩을 멘 채 한 손에 구겨진 사진을 쥐고 열린 현관 밖으로 발을 내디딘 수리영의 뒷모습.\n\nLOCATION (lock): At the rooftop home's interior entryway, crossing through the open front door from the ransacked home at dusk. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-back height one step behind 수리영, the camera completes its following move through the living room as she crosses the open entrance. Her backpack and hunched shoulders occupy the central midground, while the hand gripping the crumpled photograph hangs clearly along one side and the disturbed room recedes behind her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Entrance door (Open) — Its open inner face and doorway edges lead away from the camera; used as Exit frame directly ahead of 수리영 and endpoint of the tracking path; Ransacked living-room furnishings (Disordered) — Displaced sides and overturned forms face the following camera at irregular angles; used as Background evidence left behind as 수리영 exits; Old photograph (Crumpled, bloodstained, and removed from the wall) — The crumpled image-bearing face is only partly visible between her clenched fingers; used as Small but readable object held beside 수리영’s body; Backpack (Worn on 수리영’s back) — Its rear face is presented directly toward the following camera; used as Silhouette and load defining 수리영’s departure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience holds the interior in muted low-key contrast as 수리영 moves through the open entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 sunset interior light, open window, muddy footprints, overturned chairs, and ransacked living-room state from the reference. Preserve the young woman's clothing, backpack, and crumpled photograph, but exclude the bedroom's red circle and wall-mounted photograph from the exit 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bloodied photograph has been removed from the wall and is tightly held in Su-ri-young’s hand, while the red wall circle and the ransacked room remain behind. She leaves wearing the backpack that contains her printed maps.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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, straight-ahead wide composition with a natural wide-lens feel, holding her backpack and hunched shoulders firmly at center as the doorway forms a severe frame around her departure. Keep generous depth of field so the crumpled photograph at her side, the entrance edges, and the irregularly angled disorder receding behind her remain materially legible. Let the subdued dusk from ahead and the existing window ambience separate her darkened back from the exit without turning her into a featureless 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.\n\nSHOT TEXT (authoritative, Korean): 백팩을 멘 채 한 손에 구겨진 사진을 쥐고 열린 현관 밖으로 발을 내디딘 수리영의 뒷모습.\n\nLOCATION (lock): At the rooftop home's interior entryway, crossing through the open front door from the ransacked home at dusk. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At upper-back height one step behind 수리영, the camera completes its following move through the living room as she crosses the open entrance. Her backpack and hunched shoulders occupy the central midground, while the hand gripping the crumpled photograph hangs clearly along one side and the disturbed room recedes behind her.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Entrance door (Open) — Its open inner face and doorway edges lead away from the camera; used as Exit frame directly ahead of 수리영 and endpoint of the tracking path; Ransacked living-room furnishings (Disordered) — Displaced sides and overturned forms face the following camera at irregular angles; used as Background evidence left behind as 수리영 exits; Old photograph (Crumpled, bloodstained, and removed from the wall) — The crumpled image-bearing face is only partly visible between her clenched fingers; used as Small but readable object held beside 수리영’s body; Backpack (Worn on 수리영’s back) — Its rear face is presented directly toward the following camera; used as Silhouette and load defining 수리영’s departure.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience holds the interior in muted low-key contrast as 수리영 moves through the open entrance.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 sunset interior light, open window, muddy footprints, overturned chairs, and ransacked living-room state from the reference. Preserve the young woman's clothing, backpack, and crumpled photograph, but exclude the bedroom's red circle and wall-mounted photograph from the exit 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bloodied photograph has been removed from the wall and is tightly held in Su-ri-young’s hand, while the red wall circle and the ransacked room remain behind. She leaves wearing the backpack that contains her printed maps.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 wide shot with a gently compressed lens feel and selective focus concentrated across her upper back, backpack, clenched hand, and the partly exposed face of the crumpled photograph. Allow the ransacked interior to fall away in softer, overlapping layers while its overturned and displaced forms remain unmistakable around the route to the open door. Shape the low dusk ambience with delicate edge separation on fabric, fingers, and wrinkled paper, emphasizing the weight and finality of the step outward.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": 2 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "두 후보 모두 캐릭터가 방에서 나가는(exiting) 대신 밖에서 방 안으로 들어가는(entering) 방향으로 공간을 뒤집어 렌더링하는 치명적인 오류(Hard Violation)를 범했으나, A가 카메라의 중앙 후면 구도, 구겨진 사진의 크기, 백팩의 지도 디테일 등을 더 잘 살렸습니다." }, { "label": "B", "score": 2, "verdict_ko": "캐릭터가 밖에서 방 안을 향해 서 있어 '현관 밖으로 발을 내디딘'이라는 동선 지시를 완전히 어겼으며(Hard Violation), 손에 든 사진이 구겨지지 않고 비현실적으로 크게 렌더링되었습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S25sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 낡은 어린아이 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "거실에서 현관 밖으로 나가는 프롬프트 설정과 달리, 현관 밖에서 집 안 거실 쪽을 향해 들어가는 방향으로 잘못 연출되었습니다.", "fix_en": "The camera position and character direction are reversed (entering instead of exiting), which requires completely rebuilding the scene from an interior viewpoint.", "unfixable": true }, { "issue_ko": "손에 든 사진이 평평하게 펴져 있으며, 피가 묻어 있거나 주먹을 꽉 쥔 형태가 아닙니다.", "fix_en": "Modify the hand to a clenched fist and replace the flat photograph with a crumpled, bloodstained version.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "issue_ko": "카메라가 현관 바깥에서 실내를 바라보고 있어, 실내에서 한 걸음 뒤따라가며 수리영이 열린 문 밖으로 나가는 구도와 정반대입니다. 어질러진 방도 그녀 뒤로 멀어지는 대신 앞쪽에 펼쳐집니다.", "fix_en": "Preserve the current photograph; the reversed interior-to-exterior viewpoint and depth order cannot be corrected in place.", "unfixable": true }, { "issue_ko": "수리영이 전경에서 화면 높이 대부분을 차지하고 다리 아래가 잘려, 중앙 미드그라운드의 전신을 담는 와이드 숏보다 지나치게 타이트합니다.", "fix_en": "Expand the frame around the existing image into a wider composition that includes her full stepping figure while keeping her centered.", "unfixable": false }, { "issue_ko": "계약에서 요구한 웅크린 어깨와 달리 수리영의 어깨와 상체가 비교적 곧게 서 있습니다.", "fix_en": "Lower and round her shoulders slightly and add a subtle forward hunch while preserving her position and movement axis.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Modify the hand to a clenched fist and replace the flat photograph with a crumpled, bloodstained version.\n- Expand the frame around the existing image into a wider composition that includes her full stepping figure while keeping her centered.\n- Lower and round her shoulders slightly and add a subtle forward hunch while preserving her position and movement axis.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "피사체가 집 밖(퇴장)이 아닌 안으로 향하는 동선 오류(카메라 위치 위반)가 있으나, 신체와 소품 묘사가 비교적 온전하여 A보다 우위에 있습니다." }, { "label": "A", "score": 2, "verdict_ko": "B와 동일한 동선 및 카메라 위치 오류를 범한 데다, 사진을 쥔 오른손에 손가락이 기형적으로 중복 생성되는 심각한 신체 오류(Hard Violation)가 있습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 4, "verdict_ko": "피사체가 집 밖(퇴장)이 아닌 안으로 향하는 동선 오류(카메라 위치 위반)가 있으나, 신체와 소품 묘사가 비교적 온전하여 A보다 우위에 있습니다." }, { "label": "A", "score": 2, "verdict_ko": "B와 동일한 동선 및 카메라 위치 오류를 범한 데다, 사진을 쥔 오른손에 손가락이 기형적으로 중복 생성되는 심각한 신체 오류(Hard Violation)가 있습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 5, "verdict_ko": "두 이미지 모두 거실이 캐릭터 앞에 배치되어 카메라 방향 연출에 실패했으나, B는 레퍼런스와 일치하는 사진 소품 묘사와 안정적인 손 묘사를 보여주어 더 낫습니다." }, { "label": "A", "score": 4, "verdict_ko": "카메라 방향 지시를 어겼을 뿐만 아니라, 사진을 쥔 손의 형태가 심하게 무너지고 사진 안의 피사체도 전혀 알아볼 수 없습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "두 이미지 모두 거실이 캐릭터 앞에 배치되어 카메라 방향 연출에 실패했으나, B는 레퍼런스와 일치하는 사진 소품 묘사와 안정적인 손 묘사를 보여주어 더 낫습니다." }, { "label": "B", "score": 4, "verdict_ko": "카메라 방향 지시를 어겼을 뿐만 아니라, 사진을 쥔 손의 형태가 심하게 무너지고 사진 안의 피사체도 전혀 알아볼 수 없습니다." } ] }, "combined": { "totals": { "A": 7, "B": 8 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S25sh1" } }, "groupbg::안개 낀 포구": { "input_fingerprint": "a60d7781ebc2c347", "meta": { "model": "gpt-image-2", "size": "1536x864", "pack": "11.202607220237", "contract": "bgfirst_full_v3", "group_sig": { "key": "안개 낀 포구", "tags": [ "S26sh1", "S26sh3" ] }, "context_sig": "a4aab31ece0f8d75" }, "prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: On shoreline rocks at the edge of an empty harbor, surrounded by dense sea fog at dusk.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n안개 낀 포구와 갯바위: 시야를 가리는 짙은 안개로 덮인 텅 빈 포구와 거친 질감의 바위지대다. (특징: 시야를 가리는 자욱한 안개; 물기 젖은 거친 갯바위; 선착장 고정용 구조물)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#26. 포구 / 해질 무렵\n- 텅 빈 포구. 안개로 인해 배는 보이지 않고.\n- 그때 안개를 뚫고 배 한척이 다가오는 게 보인다.\n\nTIME OF DAY (lock): dusk, foggy after rain.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "effective_prompt": "Create ONE empty live-action location background photograph — NO PEOPLE, no figures, no body parts, no silhouettes, no shadows or reflections of people anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods and produce in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch of a shot that happens at this location: use it ONLY as spatial evidence — what this place contains, how its ground, structures and landmarks are arranged and proportioned. Ignore the sketched people and arrows entirely, and do NOT copy its line style: render a fully photographic, physically plausible real place that fits THE LOCATION text below.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nTHE LOCATION — Incheon, South Korea, 2026; all people are Korean unless stated: On shoreline rocks at the edge of an empty harbor, surrounded by dense sea fog at dusk.\n\nLOCATION DETAIL (descriptive evidence about this place from the production's location records — evidence, not a staging order): use it to understand what kind of place this is and what it permanently contains. Carry over ONLY its enduring physical features — layout, structures, ground and materials, fixed equipment and installations. Do NOT treat any lighting, weather or time-of-day wording, momentary object states, depicted events or subjective impressions in it as fixed facts of the place; the TIME OF DAY lock below is the sole authority for time and lighting.\n안개 낀 포구와 갯바위: 시야를 가리는 짙은 안개로 덮인 텅 빈 포구와 거친 질감의 바위지대다. (특징: 시야를 가리는 자욱한 안개; 물기 젖은 거친 갯바위; 선착장 고정용 구조물)\n\nSCENE EVIDENCE (verbatim quotes from the screenplay about this place — treat them as evidence of what the location physically contains and looks like; stage the PLACE those moments happen in, but do NOT depict the momentary actions, people or staged props themselves):\n- S#26. 포구 / 해질 무렵\n- 텅 빈 포구. 안개로 인해 배는 보이지 않고.\n- 그때 안개를 뚫고 배 한척이 다가오는 게 보인다.\n\nTIME OF DAY (lock): dusk, foggy after rain.\n\nRender ONE photorealistic empty location photograph, 16:9, neutral enough that every shot of this place can be staged from it later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_안개_낀_포구_a58666.png", "asset_id": "3049c2cf-9a41-44fe-906c-731726d0142b", "input_asset_ids": [ "12031eae-9c14-4aac-8bc1-708a76886871" ], "origin_tag": "S26sh1", "place_text": "On shoreline rocks at the edge of an empty harbor, surrounded by dense sea fog at dusk.", "origin_inputs": { "place_text": "On shoreline rocks at the edge of an empty harbor, surrounded by dense sea fog at dusk.", "time_of_day_en": "dusk, foggy after rain", "conti_asset_id": "12031eae-9c14-4aac-8bc1-708a76886871" } }, "S26sh1::bgfirst_bg": { "input_fingerprint": "905eb16912f0b046", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 짙은 안개가 깔린 포구의 갯바위 위, 무릎을 감싸 안은 채 바다를 주시하는 수리영의 전신.\n\nLOCATION (lock): On shoreline rocks at the edge of an empty harbor, surrounded by dense sea fog at dusk.\n\nTIME OF DAY (lock): dusk, foggy after rain.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height behind and slightly to one side, a static wide frame places 수리영’s full seated figure off-center against the fog-covered sea. Her compact posture anchors the foreground rocks while the empty port and obscured water create open negative space along her fixed line of sight.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Sea (Covered by dense fog); used as Broad negative space receiving 수리영’s distant gaze; Fog (Dense and moving in from the sea); used as Atmospheric layer obscuring vessels and reducing spatial depth; Shore rocks (Occupied by 수리영); used as Foreground support and scale reference beneath 수리영; Port (Empty; the rain has stopped); used as Environmental context emphasizing isolation.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Soft dusk ambience is diffused by the stated dense fog, producing restrained contrast and a muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 짙은 안개가 깔린 포구의 갯바위 위, 무릎을 감싸 안은 채 바다를 주시하는 수리영의 전신.\n\nLOCATION (lock): On shoreline rocks at the edge of an empty harbor, surrounded by dense sea fog at dusk.\n\nTIME OF DAY (lock): dusk, foggy after rain.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height behind and slightly to one side, a static wide frame places 수리영’s full seated figure off-center against the fog-covered sea. Her compact posture anchors the foreground rocks while the empty port and obscured water create open negative space along her fixed line of sight.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Sea (Covered by dense fog); used as Broad negative space receiving 수리영’s distant gaze; Fog (Dense and moving in from the sea); used as Atmospheric layer obscuring vessels and reducing spatial depth; Shore rocks (Occupied by 수리영); used as Foreground support and scale reference beneath 수리영; Port (Empty; the rain has stopped); used as Environmental context emphasizing isolation.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Soft dusk ambience is diffused by the stated dense fog, producing restrained contrast and a muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S26sh1__bgfirst_bg.png", "asset_id": "3919a824-7e0f-4e57-8f02-130e424e0106", "input_asset_ids": [ "12031eae-9c14-4aac-8bc1-708a76886871", "3049c2cf-9a41-44fe-906c-731726d0142b" ] }, "S26sh1": { "input_fingerprint": "945d88385728cebe", "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, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 짙은 안개가 깔린 포구의 갯바위 위, 무릎을 감싸 안은 채 바다를 주시하는 수리영의 전신.\n\nLOCATION (lock): On shoreline rocks at the edge of an empty harbor, surrounded by dense sea fog at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height behind and slightly to one side, a static wide frame places 수리영’s full seated figure off-center against the fog-covered sea. Her compact posture anchors the foreground rocks while the empty port and obscured water create open negative space along her fixed line of sight.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Sea (Covered by dense fog); used as Broad negative space receiving 수리영’s distant gaze; Fog (Dense and moving in from the sea); used as Atmospheric layer obscuring vessels and reducing spatial depth; Shore rocks (Occupied by 수리영); used as Foreground support and scale reference beneath 수리영; Port (Empty; the rain has stopped); used as Environmental context emphasizing isolation.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Soft dusk ambience is diffused by the stated dense fog, producing restrained contrast and a muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young still has her phone, the bloodied old photograph, and her backpack containing the printed maps while she waits at the harbor.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 짙은 안개가 깔린 포구의 갯바위 위, 무릎을 감싸 안은 채 바다를 주시하는 수리영의 전신.\n\nLOCATION (lock): On shoreline rocks at the edge of an empty harbor, surrounded by dense sea fog at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height behind and slightly to one side, a static wide frame places 수리영’s full seated figure off-center against the fog-covered sea. Her compact posture anchors the foreground rocks while the empty port and obscured water create open negative space along her fixed line of sight.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Sea (Covered by dense fog); used as Broad negative space receiving 수리영’s distant gaze; Fog (Dense and moving in from the sea); used as Atmospheric layer obscuring vessels and reducing spatial depth; Shore rocks (Occupied by 수리영); used as Foreground support and scale reference beneath 수리영; Port (Empty; the rain has stopped); used as Environmental context emphasizing isolation.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Soft dusk ambience is diffused by the stated dense fog, producing restrained contrast and a muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young still has her phone, the bloodied old photograph, and her backpack containing the printed maps while she waits at the harbor.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "B": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 짙은 안개가 깔린 포구의 갯바위 위, 무릎을 감싸 안은 채 바다를 주시하는 수리영의 전신.\n\nLOCATION (lock): On shoreline rocks at the edge of an empty harbor, surrounded by dense sea fog at dusk. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height behind and slightly to one side, a static wide frame places 수리영’s full seated figure off-center against the fog-covered sea. Her compact posture anchors the foreground rocks while the empty port and obscured water create open negative space along her fixed line of sight.\n- FRAMING SCALE: wide shot\n- KEY BACKGROUND ELEMENTS: Sea (Covered by dense fog); used as Broad negative space receiving 수리영’s distant gaze; Fog (Dense and moving in from the sea); used as Atmospheric layer obscuring vessels and reducing spatial depth; Shore rocks (Occupied by 수리영); used as Foreground support and scale reference beneath 수리영; Port (Empty; the rain has stopped); used as Environmental context emphasizing isolation.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Soft dusk ambience is diffused by the stated dense fog, producing restrained contrast and a muted-neutral palette.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young still has her phone, the bloodied old photograph, and her backpack containing the printed maps while she waits at the harbor.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S26sh1__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S26sh1.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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_안개_낀_포구_a58666.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" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "바위 위에 놓인 지도와 사진이 지형을 무시하고 평면적인 스티커처럼 처리되어 사실성이 떨어지며, 지도가 가방 안에 있어야 한다는 설정과도 어긋납니다." }, { "label": "B", "score": 7, "verdict_ko": "안개 낀 항구의 배경과 레퍼런스 위치를 정확히 반영했으며, 구도와 인물의 자세, 가방 안에 수납된 소지품의 물리적 묘사가 자연스럽습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "바위 위에 놓인 지도와 사진이 지형을 무시하고 평면적인 스티커처럼 처리되어 사실성이 떨어지며, 지도가 가방 안에 있어야 한다는 설정과도 어긋납니다." }, { "label": "B", "score": 7, "verdict_ko": "안개 낀 항구의 배경과 레퍼런스 위치를 정확히 반영했으며, 구도와 인물의 자세, 가방 안에 수납된 소지품의 물리적 묘사가 자연스럽습니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "카메라 앵글, 배경, 인물의 자세가 지시문과 잘 맞으며, 피 묻은 사진과 지도 등 요구된 모든 소지품이 명확히 포함되어 우수합니다." }, { "label": "A", "score": 5, "verdict_ko": "전반적인 구도와 분위기는 부합하나, 소지품 중 명시된 '피 묻은 낡은 사진'이 보이지 않아 디테일에서 감점되었습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "카메라 앵글, 배경, 인물의 자세가 지시문과 잘 맞으며, 피 묻은 사진과 지도 등 요구된 모든 소지품이 명확히 포함되어 우수합니다." }, { "label": "B", "score": 5, "verdict_ko": "전반적인 구도와 분위기는 부합하나, 소지품 중 명시된 '피 묻은 낡은 사진'이 보이지 않아 디테일에서 감점되었습니다." } ] }, "combined": { "totals": { "A": 11, "B": 12 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 } }, "totals": { "A": 11, "B": 12 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 4, "verdict_ko": "바위 위에 놓인 지도와 사진이 지형을 무시하고 평면적인 스티커처럼 처리되어 사실성이 떨어지며, 지도가 가방 안에 있어야 한다는 설정과도 어긋납니다." }, { "label": "B", "score": 7, "verdict_ko": "안개 낀 항구의 배경과 레퍼런스 위치를 정확히 반영했으며, 구도와 인물의 자세, 가방 안에 수납된 소지품의 물리적 묘사가 자연스럽습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/groupbg_안개_낀_포구_a58666.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "인물이 무릎을 감싸 안는 자세(무릎을 감싸 안은 채)를 취하지 않고 단순히 팔을 무릎 위에 얹고 있습니다.", "fix_en": "Wrap both of her arms securely around her raised knees so she is actively hugging them to her chest.", "unfixable": false }, { "issue_ko": "지시문에 명시된 '피 묻은 오래된 사진'이 보이지 않습니다.", "fix_en": "Add a small, old, blood-stained photograph either held in her hands or visibly tucked into the backpack's outer pocket alongside the phone and map.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Wrap both of her arms securely around her raised knees so she is actively hugging them to her chest.\n- Add a small, old, blood-stained photograph either held in her hands or visibly tucked into the backpack's outer pocket alongside the phone and map.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "무릎을 감싸 안은 정확한 포즈와 배낭 곁에 배치된 피 묻은 낡은 사진 등 모든 디테일을 충실히 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "전반적인 분위기와 구도는 좋으나, 필수 소품인 피 묻은 낡은 사진이 누락되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "무릎을 감싸 안은 정확한 포즈와 배낭 곁에 배치된 피 묻은 낡은 사진 등 모든 디테일을 충실히 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "전반적인 분위기와 구도는 좋으나, 필수 소품인 피 묻은 낡은 사진이 누락되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 앵글과 배경을 잘 구현했으며, 피 묻은 사진을 포함한 모든 요구 소품이 화면에 나타납니다." }, { "label": "B", "score": 5, "verdict_ko": "카메라 구도와 배경은 훌륭하지만, 필수 소품인 '피 묻은 낡은 사진'이 완전히 누락되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 앵글과 배경을 잘 구현했으며, 피 묻은 사진을 포함한 모든 요구 소품이 화면에 나타납니다." }, { "label": "A", "score": 5, "verdict_ko": "카메라 구도와 배경은 훌륭하지만, 필수 소품인 '피 묻은 낡은 사진'이 완전히 누락되었습니다." } ] }, "combined": { "totals": { "A": 11, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S26sh1__bgfirst_bg.png", "bg_asset_id": "3919a824-7e0f-4e57-8f02-130e424e0106", "bg_record_key": "S26sh1::bgfirst_bg", "chain_winner": false, "authority": "groupbg", "group_key": "안개 낀 포구", "groupbg_asset_id": "3049c2cf-9a41-44fe-906c-731726d0142b" }, "ref_mode": "그룹 배경+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" } }, "S26sh3::variants": { "author_fp": "d49ddacfcf206256", "author": { "variants": [ { "approach_ko": "얕은 심도로 인우의 긴장된 상반신과 닻줄을 쥔 손의 물리적 힘을 밀착해 강조한다.", "prompt_en": "Use a restrained, intimate medium-shot treatment with a gently compressed lens feel and shallow depth of field. Hold crisp focus across In-woo’s face, braced upper body, and tightly engaged hands, emphasizing the rope’s coarse wet fibers and the strain carried through his arms; let the diagonal line and docking edge recede softly into dense fog. Shape the diffused dusk light subtly from the open-harbor side, with quiet wet highlights on skin, fabric, rope, and the boat’s worn surfaces." }, { "approach_ko": "넓은 공간감과 깊은 초점으로 인우와 배, 닻줄, 선착장의 기하학적 관계를 선명하게 드러낸다.", "prompt_en": "Photograph the medium frame with a broader, more spatial lens feel and comparatively deep focus, making the physical geometry of the docking maneuver immediately legible. Use In-woo’s off-center upper body as the near visual anchor while the taut diagonal rope, oblique boat side, and angled dock edge form layered lines into the fogbound harbor. Keep the fog-diffused dusk illumination broad and low-key, allowing subdued reflections from rain-darkened surfaces to separate each depth plane without dramatized highlights." } ] }, "reused": false }, "S26sh3": { "input_fingerprint": "eb823b80ee2a76c9", "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, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 짙은 안개를 가르고 다가온 작은 배 위에서 굵은 닻줄을 꽉 쥔 인우의 상반신.\n\nLOCATION (lock): At the fogbound harbor mooring, on the deck edge of a newly arrived small boat beside the dock. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside the docking point, the camera finishes tracking toward 인우 at a front-quarter angle as he braces over the mooring line on the small boat. His upper body occupies one side of the medium frame, leaving the fogged harbor and the boat’s docking edge visible along his working arm.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mooring line (Gripped and being secured) — The line runs diagonally away from 인우 toward the dock; used as Physical action line running from 인우’s hands toward the docking point; Small boat (Arrived at the docking point) — Its docking side faces the camera at an oblique angle; used as Midground platform placing 인우 above the docking edge; Fog (Dense over the harbor); used as Background depth layer isolating the arriving boat; Docking point (Empty except for the arriving boat) — Its boat-facing edge angles across the lower background; used as Spatial anchor beside the boat.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fog-diffused dusk ambience keeps 인우 and the boat in subdued, low-key neutral tones.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bloodied photograph remains in Su-ri-young’s possession, and her backpack still contains the printed maps as In-woo secures his boat.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 짙은 안개를 가르고 다가온 작은 배 위에서 굵은 닻줄을 꽉 쥔 인우의 상반신.\n\nLOCATION (lock): At the fogbound harbor mooring, on the deck edge of a newly arrived small boat beside the dock. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside the docking point, the camera finishes tracking toward 인우 at a front-quarter angle as he braces over the mooring line on the small boat. His upper body occupies one side of the medium frame, leaving the fogged harbor and the boat’s docking edge visible along his working arm.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mooring line (Gripped and being secured) — The line runs diagonally away from 인우 toward the dock; used as Physical action line running from 인우’s hands toward the docking point; Small boat (Arrived at the docking point) — Its docking side faces the camera at an oblique angle; used as Midground platform placing 인우 above the docking edge; Fog (Dense over the harbor); used as Background depth layer isolating the arriving boat; Docking point (Empty except for the arriving boat) — Its boat-facing edge angles across the lower background; used as Spatial anchor beside the boat.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fog-diffused dusk ambience keeps 인우 and the boat in subdued, low-key neutral tones.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bloodied photograph remains in Su-ri-young’s possession, and her backpack still contains the printed maps as In-woo secures his boat.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, intimate medium-shot treatment with a gently compressed lens feel and shallow depth of field. Hold crisp focus across In-woo’s face, braced upper body, and tightly engaged hands, emphasizing the rope’s coarse wet fibers and the strain carried through his arms; let the diagonal line and docking edge recede softly into dense fog. Shape the diffused dusk light subtly from the open-harbor side, with quiet wet highlights on skin, fabric, rope, and the boat’s worn surfaces.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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, foggy after rain.\n\nSHOT TEXT (authoritative, Korean): 짙은 안개를 가르고 다가온 작은 배 위에서 굵은 닻줄을 꽉 쥔 인우의 상반신.\n\nLOCATION (lock): At the fogbound harbor mooring, on the deck edge of a newly arrived small boat beside the dock. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height beside the docking point, the camera finishes tracking toward 인우 at a front-quarter angle as he braces over the mooring line on the small boat. His upper body occupies one side of the medium frame, leaving the fogged harbor and the boat’s docking edge visible along his working arm.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mooring line (Gripped and being secured) — The line runs diagonally away from 인우 toward the dock; used as Physical action line running from 인우’s hands toward the docking point; Small boat (Arrived at the docking point) — Its docking side faces the camera at an oblique angle; used as Midground platform placing 인우 above the docking edge; Fog (Dense over the harbor); used as Background depth layer isolating the arriving boat; Docking point (Empty except for the arriving boat) — Its boat-facing edge angles across the lower background; used as Spatial anchor beside the boat.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Fog-diffused dusk ambience keeps 인우 and the boat in subdued, low-key neutral tones.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The bloodied photograph remains in Su-ri-young’s possession, and her backpack still contains the printed maps as In-woo secures his boat.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nPhotograph the medium frame with a broader, more spatial lens feel and comparatively deep focus, making the physical geometry of the docking maneuver immediately legible. Use In-woo’s off-center upper body as the near visual anchor while the taut diagonal rope, oblique boat side, and angled dock edge form layered lines into the fogbound harbor. Keep the fog-diffused dusk illumination broad and low-key, allowing subdued reflections from rain-darkened surfaces to separate each depth plane without dramatized 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." }, "totals": { "A": 7, "B": 4 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 카메라 위치(정박지 옆)를 잘 지켰으며, 레퍼런스와 일치하는 콘크리트 부두와 안개 낀 바위 배경을 충실하게 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "카메라가 정박지 옆이 아닌 배 쪽에 위치해 있으며, 레퍼런스에 없는 부유식 폰툰과 조명을 배치해 장소 제약을 위반했고 비가 내리고 있어 아쉽습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S26sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "밧줄을 쥐고 있는 인우의 왼쪽 손(화면 앞쪽)의 손가락 구조가 심하게 뭉개져 밧줄과 융합되어 있습니다.", "fix_en": "Redraw the extended left hand to have a clear, anatomically correct thumb and fingers naturally wrapping around the rope.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw the extended left hand to have a clear, anatomically correct thumb and fingers naturally wrapping around the rope.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "인우의 외모, 의상, 안개 낀 젖은 항구의 분위기와 지정된 미디엄 샷 프레이밍은 완벽히 구현되었으나, 우측 선창의 닻줄이 금속 기둥을 물리적으로 관통하는 심각한 구조적 오류(Hard Violation)가 있습니다." }, { "label": "B", "score": 5, "verdict_ko": "A와 시각적으로 동일한 이미지를 생성하였으며, 프롬프트의 인물 및 배경 조건은 훌륭하게 충족했으나 선창 기둥을 관통하는 닻줄의 형태적 불가능성이 동일하게 나타납니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "인우의 외모, 의상, 안개 낀 젖은 항구의 분위기와 지정된 미디엄 샷 프레이밍은 완벽히 구현되었으나, 우측 선창의 닻줄이 금속 기둥을 물리적으로 관통하는 심각한 구조적 오류(Hard Violation)가 있습니다." }, { "label": "B", "score": 5, "verdict_ko": "A와 시각적으로 동일한 이미지를 생성하였으며, 프롬프트의 인물 및 배경 조건은 훌륭하게 충족했으나 선창 기둥을 관통하는 닻줄의 형태적 불가능성이 동일하게 나타납니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 샷의 구도와 인물의 동작을 충실히 구현했으며, 밧줄을 쥔 양손의 해부학적 형태가 자연스럽습니다." }, { "label": "B", "score": 3, "verdict_ko": "아래쪽 손(오른손)의 우측에 기형적인 손가락 형태가 생성되어 해부학적 오류(Hard Violation)가 발생했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 샷의 구도와 인물의 동작을 충실히 구현했으며, 밧줄을 쥔 양손의 해부학적 형태가 자연스럽습니다." }, { "label": "A", "score": 3, "verdict_ko": "아래쪽 손(오른손)의 우측에 기형적인 손가락 형태가 생성되어 해부학적 오류(Hard Violation)가 발생했습니다." } ] }, "combined": { "totals": { "A": 8, "B": 12 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S26sh1" } }, "S27sh1::variants": { "author_fp": "ad87ce78f97623c3", "author": { "variants": [ { "approach_ko": "얕은 심도로 혜수의 충격받은 표정을 밀착해 포착하고, 난장판과 출입문은 흐릿하지만 식별 가능한 층으로 남긴다.", "prompt_en": "Use an intimate, gently compressed medium-shot lens feel, holding Hye-su’s widened eyes and the phone-hand contact in crisp focus while the wrecked room recedes into restrained, legible softness. Place her slightly off-center toward the unseen detective, leaving the entrance area compressed against the far edge; let soft dusk illumination graze one side of her mature face while the opposite side falls into controlled low-key shadow. Keep the performance tense and arrested, with minute facial strain rather than exaggerated expression." }, { "approach_ko": "깊은 초점의 환경적 미디엄 숏으로 혜수와 뒤편의 난장판, 가장자리 출입문을 하나의 불안한 공간 구조로 묶는다.", "prompt_en": "Use a more spatial, deep-focus medium-shot treatment, keeping Hye-su dominant while the overturned furnishings, muddy traces, and peripheral entrance remain clearly readable at their true distance and scale. Build an uneasy layered composition with her upper body anchoring the foreground, diagonal disorder crossing the background, and the entrance held at the extreme edge as a subtle visual pull. Render the dusk interior with broad, subdued tonal separation and muted neutral color, favoring environmental tension over facial isolation." } ] }, "reused": false }, "S27sh1": { "input_fingerprint": "002b62e5eef66dfd", "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): 어둑하고 난장판이 된 옥탑방 거실, 휴대폰을 귀에 댄 채 두 눈을 크게 뜬 혜수의 상반신.\n\nLOCATION (lock): The dark, ransacked living room inside the rooftop home at dusk, near the rattling entrance door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height in front of 혜수, a static medium-close composition holds her widened eyes and the phone pressed to her ear, with her body turned slightly toward the unseen 김 형사. The ransacked living room stays legible behind her, while the entrance area remains at the far edge in preparation for the later lateral move.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mobile phone (Playing or displaying 수리영’s recorded message) — Its screen face is turned toward 혜수 and away from the camera; used as Handheld story prop held against 혜수’s face while she checks the recorded message; Living-room furnishings (Ransacked) — Displaced and overturned sides face the camera behind her; used as Background evidence surrounding 혜수’s anxious reaction; Entrance door (Moving audibly in the wind) — Its interior-facing side is visible near the edge of frame; used as Peripheral background endpoint for the scene’s later turn toward the sound.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The dim dusk interior is rendered with restrained low-key contrast and muted neutral color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dim sunset light, muddy footprints, overturned furniture, open window, and ransacked living-room materials from the reference. Exclude the younger woman, her backpack, and the removed photograph, and do not import the separate bedroom's red-circle imagery.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with the muddy footprints, open living-room window, and overturned chairs left in place. The bedroom’s red circle remains, but its bloodied photograph has been taken by Su-ri-young.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 어둑하고 난장판이 된 옥탑방 거실, 휴대폰을 귀에 댄 채 두 눈을 크게 뜬 혜수의 상반신.\n\nLOCATION (lock): The dark, ransacked living room inside the rooftop home at dusk, near the rattling entrance door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height in front of 혜수, a static medium-close composition holds her widened eyes and the phone pressed to her ear, with her body turned slightly toward the unseen 김 형사. The ransacked living room stays legible behind her, while the entrance area remains at the far edge in preparation for the later lateral move.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mobile phone (Playing or displaying 수리영’s recorded message) — Its screen face is turned toward 혜수 and away from the camera; used as Handheld story prop held against 혜수’s face while she checks the recorded message; Living-room furnishings (Ransacked) — Displaced and overturned sides face the camera behind her; used as Background evidence surrounding 혜수’s anxious reaction; Entrance door (Moving audibly in the wind) — Its interior-facing side is visible near the edge of frame; used as Peripheral background endpoint for the scene’s later turn toward the sound.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The dim dusk interior is rendered with restrained low-key contrast and muted neutral color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dim sunset light, muddy footprints, overturned furniture, open window, and ransacked living-room materials from the reference. Exclude the younger woman, her backpack, and the removed photograph, and do not import the separate bedroom's red-circle imagery.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with the muddy footprints, open living-room window, and overturned chairs left in place. The bedroom’s red circle remains, but its bloodied photograph has been taken by Su-ri-young.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse an intimate, gently compressed medium-shot lens feel, holding Hye-su’s widened eyes and the phone-hand contact in crisp focus while the wrecked room recedes into restrained, legible softness. Place her slightly off-center toward the unseen detective, leaving the entrance area compressed against the far edge; let soft dusk illumination graze one side of her mature face while the opposite side falls into controlled low-key shadow. Keep the performance tense and arrested, with minute facial strain rather than exaggerated 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): dusk.\n\nSHOT TEXT (authoritative, Korean): 어둑하고 난장판이 된 옥탑방 거실, 휴대폰을 귀에 댄 채 두 눈을 크게 뜬 혜수의 상반신.\n\nLOCATION (lock): The dark, ransacked living room inside the rooftop home at dusk, near the rattling entrance door. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At chest height in front of 혜수, a static medium-close composition holds her widened eyes and the phone pressed to her ear, with her body turned slightly toward the unseen 김 형사. The ransacked living room stays legible behind her, while the entrance area remains at the far edge in preparation for the later lateral move.\n- FRAMING SCALE: medium shot\n- KEY BACKGROUND ELEMENTS: Mobile phone (Playing or displaying 수리영’s recorded message) — Its screen face is turned toward 혜수 and away from the camera; used as Handheld story prop held against 혜수’s face while she checks the recorded message; Living-room furnishings (Ransacked) — Displaced and overturned sides face the camera behind her; used as Background evidence surrounding 혜수’s anxious reaction; Entrance door (Moving audibly in the wind) — Its interior-facing side is visible near the edge of frame; used as Peripheral background endpoint for the scene’s later turn toward the sound.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The dim dusk interior is rendered with restrained low-key contrast and muted neutral color.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same dim sunset light, muddy footprints, overturned furniture, open window, and ransacked living-room materials from the reference. Exclude the younger woman, her backpack, and the removed photograph, and do not import the separate bedroom's red-circle imagery.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The rooftop room remains ransacked, with the muddy footprints, open living-room window, and overturned chairs left in place. The bedroom’s red circle remains, but its bloodied photograph has been taken by Su-ri-young.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 혜수 (한국인 40대 중반 여성, 성숙한 얼굴선, 짙은 검은색 머리카락, 단정한 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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 more spatial, deep-focus medium-shot treatment, keeping Hye-su dominant while the overturned furnishings, muddy traces, and peripheral entrance remain clearly readable at their true distance and scale. Build an uneasy layered composition with her upper body anchoring the foreground, diagonal disorder crossing the background, and the entrance held at the extreme edge as a subtle visual pull. Render the dusk interior with broad, subdued tonal separation and muted neutral color, favoring environmental tension over facial 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": { "A": 7, "B": 6 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "캐릭터 레퍼런스의 제복 의상을 누락했지만, 휴대폰 화면이 카메라 반대편을 향해야 한다는 명시적인 방향 지시와 구도를 정확히 준수했습니다." }, { "label": "B", "score": 6, "verdict_ko": "경찰 제복은 훌륭하게 재현했으나, 휴대폰 화면이 카메라를 향하고 있어 프롬프트의 핵심 소품 방향 지시를 정면으로 위반했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S25sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 혜수: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 스마트폰: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "기준 이미지와 달리 뒤쪽 벽의 열린 현관문이 닫힌 일반 문으로 바뀌었으며, 원래의 철제 현관문은 우측 전경에 잘못 배치되어 공간 구조가 일치하지 않습니다.", "fix_en": "Remove the heavy metal door on the right edge and replace the closed door on the back wall with the open heavy entrance door matching the reference.", "unfixable": true }, { "issue_ko": "바닥의 흙 묻은 발자국 방향이 기준 이미지(방 안쪽을 향함)와 반대로 카메라 쪽을 향하고 있습니다.", "fix_en": "Redraw the muddy footprints on the floor so they point away from the camera toward the back wall.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Minimize the centrally placed entrance door and retain only a narrow peripheral indication of the entrance area at the frame edge, preserving Hye-su’s scale and placement.", "issue_ko": "고정된 현관문이 프레임 중앙부 배경에 놓여 있어, 현관 구역을 화면의 먼 가장자리에 압축해 두라는 구도와 다릅니다. 오른쪽 가장자리의 문도 전경에서 지나치게 크게 보여 주변적 배경 요소가 되지 못합니다.", "unfixable": true } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw the muddy footprints on the floor so they point away from the camera toward the back wall.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "레퍼런스 이미지에 나타난 맨발 형태의 진흙 발자국을 정확히 유지하였으며, 요구된 앵글과 인물의 표정 연기를 훌륭하게 구현했습니다." }, { "label": "B", "score": 7, "verdict_ko": "대부분의 지시를 잘 따랐으나, 레퍼런스의 맨발 발자국을 신발 밑창 패턴이 있는 발자국으로 임의 변형하여 연속성 디테일에서 감점되었습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "레퍼런스 이미지에 나타난 맨발 형태의 진흙 발자국을 정확히 유지하였으며, 요구된 앵글과 인물의 표정 연기를 훌륭하게 구현했습니다." }, { "label": "B", "score": 7, "verdict_ko": "대부분의 지시를 잘 따랐으나, 레퍼런스의 맨발 발자국을 신발 밑창 패턴이 있는 발자국으로 임의 변형하여 연속성 디테일에서 감점되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 구도와 옥탑방의 난장판이 된 배경을 훌륭하게 구현했으나, 레퍼런스에 없는 현관문이 화면 우측에 중복 생성된 공간적 오류가 있습니다. 하지만 신체 구조는 정상적입니다." }, { "label": "B", "score": 3, "verdict_ko": "A와 동일한 배경과 구도 오류를 공유하는 동시에, 휴대폰을 쥔 손가락이 5개로 묘사되는 치명적인 해부학적 하드 위반이 발생하여 감점되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지시된 구도와 옥탑방의 난장판이 된 배경을 훌륭하게 구현했으나, 레퍼런스에 없는 현관문이 화면 우측에 중복 생성된 공간적 오류가 있습니다. 하지만 신체 구조는 정상적입니다." }, { "label": "A", "score": 3, "verdict_ko": "A와 동일한 배경과 구도 오류를 공유하는 동시에, 휴대폰을 쥔 손가락이 5개로 묘사되는 치명적인 해부학적 하드 위반이 발생하여 감점되었습니다." } ] }, "combined": { "totals": { "A": 12, "B": 14 }, "ranking": [ "B", "A" ], "agreement": false, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S25sh1" } }, "S28sh3::bgfirst_bg": { "input_fingerprint": "2f838a442017250f", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 수리영의 손이 구겨지고 피 묻은 낡은 사진을 인우의 눈앞으로 바짝 들이민 순간의 클로즈업.\n\nLOCATION (lock): On the exposed deck of a small fishing boat underway at sea in the fading dusk light.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height between the two characters, the camera completes its push toward 수리영’s extended hand, keeping the crumpled photograph in the central foreground and 인우’s alarmed eyes sharply readable beyond it. The photograph occupies roughly a third of the frame, while a sliver of 수리영’s shoulder and 인우’s recoiling upper body preserve the confrontation’s scale and axis.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Old photograph (Crumpled, worn, and bloodstained) — Its worn image-bearing front faces both 인우 and the camera, with the rural restaurant image and small 금월 writing visible; used as Central foreground evidence linking 수리영’s accusation to 인우’s reaction; Boat deck (At sea) — The deck lines recede behind 인우 at an oblique angle; used as Soft background context retaining the confrontation’s location; Sea (Surrounding the small fishing boat); used as Distant environmental layer beyond the deck.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience keeps the deck confrontation in muted low-key tones, with the photograph’s existing blood as a deep red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 수리영의 손이 구겨지고 피 묻은 낡은 사진을 인우의 눈앞으로 바짝 들이민 순간의 클로즈업.\n\nLOCATION (lock): On the exposed deck of a small fishing boat underway at sea in the fading dusk light.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height between the two characters, the camera completes its push toward 수리영’s extended hand, keeping the crumpled photograph in the central foreground and 인우’s alarmed eyes sharply readable beyond it. The photograph occupies roughly a third of the frame, while a sliver of 수리영’s shoulder and 인우’s recoiling upper body preserve the confrontation’s scale and axis.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Old photograph (Crumpled, worn, and bloodstained) — Its worn image-bearing front faces both 인우 and the camera, with the rural restaurant image and small 금월 writing visible; used as Central foreground evidence linking 수리영’s accusation to 인우’s reaction; Boat deck (At sea) — The deck lines recede behind 인우 at an oblique angle; used as Soft background context retaining the confrontation’s location; Sea (Surrounding the small fishing boat); used as Distant environmental layer beyond the deck.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience keeps the deck confrontation in muted low-key tones, with the photograph’s existing blood as a deep red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S28sh3__bgfirst_bg.png", "asset_id": "342a221a-b48f-4ca9-afe8-e27e766a43e9", "input_asset_ids": [ "61fb473d-683e-4576-8fc5-e143b3c17928", "19fedf71-e4e7-41b0-94fb-4c3838a1d7ab" ] }, "S28sh3": { "input_fingerprint": "87927ed92ec850f9", "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): 수리영의 손이 구겨지고 피 묻은 낡은 사진을 인우의 눈앞으로 바짝 들이민 순간의 클로즈업.\n\nLOCATION (lock): On the exposed deck of a small fishing boat underway at sea in the fading dusk light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height between the two characters, the camera completes its push toward 수리영’s extended hand, keeping the crumpled photograph in the central foreground and 인우’s alarmed eyes sharply readable beyond it. The photograph occupies roughly a third of the frame, while a sliver of 수리영’s shoulder and 인우’s recoiling upper body preserve the confrontation’s scale and axis.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Old photograph (Crumpled, worn, and bloodstained) — Its worn image-bearing front faces both 인우 and the camera, with the rural restaurant image and small 금월 writing visible; used as Central foreground evidence linking 수리영’s accusation to 인우’s reaction; Boat deck (At sea) — The deck lines recede behind 인우 at an oblique angle; used as Soft background context retaining the confrontation’s location; Sea (Surrounding the small fishing boat); used as Distant environmental layer beyond the deck.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience keeps the deck confrontation in muted low-key tones, with the photograph’s existing blood as a deep red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young still has the crumpled, bloodied old photograph and holds it up to In-woo; her backpack retains the remaining printed maps.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 수리영의 손이 구겨지고 피 묻은 낡은 사진을 인우의 눈앞으로 바짝 들이민 순간의 클로즈업.\n\nLOCATION (lock): On the exposed deck of a small fishing boat underway at sea in the fading dusk light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height between the two characters, the camera completes its push toward 수리영’s extended hand, keeping the crumpled photograph in the central foreground and 인우’s alarmed eyes sharply readable beyond it. The photograph occupies roughly a third of the frame, while a sliver of 수리영’s shoulder and 인우’s recoiling upper body preserve the confrontation’s scale and axis.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Old photograph (Crumpled, worn, and bloodstained) — Its worn image-bearing front faces both 인우 and the camera, with the rural restaurant image and small 금월 writing visible; used as Central foreground evidence linking 수리영’s accusation to 인우’s reaction; Boat deck (At sea) — The deck lines recede behind 인우 at an oblique angle; used as Soft background context retaining the confrontation’s location; Sea (Surrounding the small fishing boat); used as Distant environmental layer beyond the deck.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience keeps the deck confrontation in muted low-key tones, with the photograph’s existing blood as a deep red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young still has the crumpled, bloodied old photograph and holds it up to In-woo; her backpack retains the remaining printed maps.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "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): 수리영의 손이 구겨지고 피 묻은 낡은 사진을 인우의 눈앞으로 바짝 들이민 순간의 클로즈업.\n\nLOCATION (lock): On the exposed deck of a small fishing boat underway at sea in the fading dusk light. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At hand height between the two characters, the camera completes its push toward 수리영’s extended hand, keeping the crumpled photograph in the central foreground and 인우’s alarmed eyes sharply readable beyond it. The photograph occupies roughly a third of the frame, while a sliver of 수리영’s shoulder and 인우’s recoiling upper body preserve the confrontation’s scale and axis.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Old photograph (Crumpled, worn, and bloodstained) — Its worn image-bearing front faces both 인우 and the camera, with the rural restaurant image and small 금월 writing visible; used as Central foreground evidence linking 수리영’s accusation to 인우’s reaction; Boat deck (At sea) — The deck lines recede behind 인우 at an oblique angle; used as Soft background context retaining the confrontation’s location; Sea (Surrounding the small fishing boat); used as Distant environmental layer beyond the deck.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Subdued dusk ambience keeps the deck confrontation in muted low-key tones, with the photograph’s existing blood as a deep red accent.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young still has the crumpled, bloodied old photograph and holds it up to In-woo; her backpack retains the remaining printed maps.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S28sh3__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S28sh3.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": "" }, { "label": "PROP REFERENCE — 피 묻은 낡은 어린아이 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/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L17B02.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": "" }, { "label": "PROP REFERENCE — 피 묻은 낡은 어린아이 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "사진이 프레임의 3분의 1을 차지하라는 지시보다는 작게 묘사되었으나, 카메라와 인우 모두에게 사진 앞면이 보이도록 각도를 조절한 자연스러운 구도와 인물의 위치를 성공적으로 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "수리영의 뒷모습이 프레임에 있으면서 동시에 사진을 든 손이 1인칭 시점(제3자의 손)처럼 따로 등장하는 치명적인 물리적/구도적 오류(Hard Violation)가 발생했습니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "사진이 프레임의 3분의 1을 차지하라는 지시보다는 작게 묘사되었으나, 카메라와 인우 모두에게 사진 앞면이 보이도록 각도를 조절한 자연스러운 구도와 인물의 위치를 성공적으로 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "수리영의 뒷모습이 프레임에 있으면서 동시에 사진을 든 손이 1인칭 시점(제3자의 손)처럼 따로 등장하는 치명적인 물리적/구도적 오류(Hard Violation)가 발생했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 클로즈업 샷 크기와 사진이 화면의 3분의 1을 차지하는 구도를 정확히 구현했으나, 사진을 쥔 손의 방향과 위치가 인물의 자세와 다소 어긋나 보입니다." }, { "label": "B", "score": 3, "verdict_ko": "화면 우측 상단 하늘에 창문 잔해 같은 구조물이 떠 있는 치명적인 오류(Hard Violation)가 발생했으며, 지정된 클로즈업 구도를 무시하고 너무 넓게 샷을 잡았습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 클로즈업 샷 크기와 사진이 화면의 3분의 1을 차지하는 구도를 정확히 구현했으나, 사진을 쥔 손의 방향과 위치가 인물의 자세와 다소 어긋나 보입니다." }, { "label": "A", "score": 3, "verdict_ko": "화면 우측 상단 하늘에 창문 잔해 같은 구조물이 떠 있는 치명적인 오류(Hard Violation)가 발생했으며, 지정된 클로즈업 구도를 무시하고 너무 넓게 샷을 잡았습니다." } ] }, "combined": { "totals": { "A": 10, "B": 10 }, "ranking": [ "A", "B" ], "agreement": false, "policy": 1 } }, "totals": { "A": 10, "B": 10 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "사진이 프레임의 3분의 1을 차지하라는 지시보다는 작게 묘사되었으나, 카메라와 인우 모두에게 사진 앞면이 보이도록 각도를 조절한 자연스러운 구도와 인물의 위치를 성공적으로 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "수리영의 뒷모습이 프레임에 있으면서 동시에 사진을 든 손이 1인칭 시점(제3자의 손)처럼 따로 등장하는 치명적인 물리적/구도적 오류(Hard Violation)가 발생했습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its architecture, materials, fixed features and lighting mood are spatial truth; stage the moment inside this place. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L17B02.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": "" }, { "label": "PROP REFERENCE — 피 묻은 낡은 어린아이 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "화면 우측 상단 하늘에 원본 배경 이미지에 있던 창틀의 잔상이 허공에 비현실적으로 떠 있습니다. 배경을 변형 없이 유지하라는 지침 위반입니다.", "fix_en": "Remove the floating window frame fragment from the top right sky, restoring the clear sky background.", "unfixable": false }, { "issue_ko": "인우의 양손(특히 화면 우측 하단에 보이는 왼손)의 손가락 마디와 형태가 뭉개지고 일그러져 해부학적으로 비정상적입니다.", "fix_en": "Correct the anatomy of In-woo's hands to show properly proportioned and structured human fingers.", "unfixable": false }, { "issue_ko": "낡은 사진 표면에 프롬프트가 필수 요소로 지시한 작은 '금월'이라는 글씨가 누락되었습니다.", "fix_en": "Add the small text '금월' to the front surface of the old photograph.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Remove the floating window frame fragment from the top right sky, restoring the clear sky background.\n- Correct the anatomy of In-woo's hands to show properly proportioned and structured human fingers.\n- Add the small text '금월' to the front surface of the old 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.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "사진 속 '금월' 텍스트를 포함하여 모든 세부 요구사항과 스케치의 구도를 충실하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "요구된 '금월' 텍스트가 사진에 없으며, 우측 상단 하늘에 정체불명의 구조물이 나타나는 심각한 오류가 있습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "사진 속 '금월' 텍스트를 포함하여 모든 세부 요구사항과 스케치의 구도를 충실하게 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "요구된 '금월' 텍스트가 사진에 없으며, 우측 상단 하늘에 정체불명의 구조물이 나타나는 심각한 오류가 있습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "요구된 구도와 조명을 훌륭하게 따랐으며, 소품인 사진에 핏자국과 '금월' 텍스트를 정확히 구현해 프롬프트 충실도가 높습니다." }, { "label": "B", "score": 3, "verdict_ko": "우측 상단 허공에 정체불명의 구조물 조각이 떠 있는 치명적인 생성 오류가 있으며, 사진에 '금월' 텍스트가 누락되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "요구된 구도와 조명을 훌륭하게 따랐으며, 소품인 사진에 핏자국과 '금월' 텍스트를 정확히 구현해 프롬프트 충실도가 높습니다." }, { "label": "A", "score": 3, "verdict_ko": "우측 상단 허공에 정체불명의 구조물 조각이 떠 있는 치명적인 생성 오류가 있으며, 사진에 '금월' 텍스트가 누락되었습니다." } ] }, "combined": { "totals": { "A": 7, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S28sh3__bgfirst_bg.png", "bg_asset_id": "342a221a-b48f-4ca9-afe8-e27e766a43e9", "bg_record_key": "S28sh3::bgfirst_bg", "chain_winner": true, "authority": "plate" }, "ref_mode": "재투영 배경+콘티+엔티티 (2택1: 체인 승)", "share_plan": { "ref_plan": "background" }, "plate_select": { "candidates": { "A": "L17B01", "B": "L17B02" }, "assigned": "L17B02", "choice": "Candidate B", "confident": false, "reason_ko": "지문은 수리영이 인우의 눈앞에 사진을 들이미는 클로즈업 숏으로, 갑판(B)인지 조타실(A)인지 구분할 수 있는 배경 단서가 없으므로 기존 할당을 유지함.", "kept": "L17B02" } }, "S28sh9::bgfirst_bg": { "input_fingerprint": "b08ca06d3c784142", "prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 조타실 안, 방향타를 쥔 인우의 팔목을 수리영이 양손으로 강하게 꽉 쥐어 제어하며 두 사람의 힘이 팽팽하게 부딪히는 찰나.\n\nLOCATION (lock): Just inside the small fishing boat's wheelhouse, at the helm where the two struggle over control at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the wheelhouse beside the helm at waist height, the camera completes a forceful push to within close range of the joined arms, using a restrained Dutch tilt to sharpen the imbalance. 수리영’s two hands clamp around 인우’s wrist in the center while his other arm and both torsos pull in opposing diagonals around the helm, preserving the physical cause of the struggle.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Helm (Held by 인우 during the struggle) — Its gripping side faces 인우, with its rim angled toward the waist-height camera; used as Central mechanical anchor around which both characters exert opposing force; Wheelhouse interior (Occupied by both struggling characters) — Its enclosing sides converge behind the two torsos and helm; used as Tight architectural context enclosing the physical clash.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural dusk ambience appropriate to the wheelhouse is held in tense low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "effective_prompt": "Create the EMPTY BACKGROUND PLATE for one film shot — NO PEOPLE, no figures, no body parts, no sketch lines, no arrows anywhere.\n\"Empty\" means no people only: KEEP the location's inherent occupants and stock that define the place — animals in an animal shelter, pen or farm, goods in a market, moored boats in a harbour — unless the shot text explicitly removes them.\nThe FIRST attached image is a thin-line storyboard sketch: use ONLY its camera angle, horizon, perspective and the placement/size of buildings and set masses — ignore the sketched people and arrows entirely. The SECOND attached image (LOCATION PHOTOGRAPH) is the real place: take its architecture, materials, signage and fixed features, and RE-PROJECT them into the sketch's camera. If the photograph's camera differs from the sketch's, the sketch's camera wins.\nHUMAN-SCALE CALIBRATION: derive every structure's true size from human-scale elements — a door ≈ 2m, a window ≈ 1–1.5m wide, one storey ≈ 2.5–3m; never inflate a small structure or shrink a large one.\n\nSHOT TEXT this background must serve (Korean): 조타실 안, 방향타를 쥔 인우의 팔목을 수리영이 양손으로 강하게 꽉 쥐어 제어하며 두 사람의 힘이 팽팽하게 부딪히는 찰나.\n\nLOCATION (lock): Just inside the small fishing boat's wheelhouse, at the helm where the two struggle over control at dusk.\n\nTIME OF DAY (lock): dusk.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the wheelhouse beside the helm at waist height, the camera completes a forceful push to within close range of the joined arms, using a restrained Dutch tilt to sharpen the imbalance. 수리영’s two hands clamp around 인우’s wrist in the center while his other arm and both torsos pull in opposing diagonals around the helm, preserving the physical cause of the struggle.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Helm (Held by 인우 during the struggle) — Its gripping side faces 인우, with its rim angled toward the waist-height camera; used as Central mechanical anchor around which both characters exert opposing force; Wheelhouse interior (Occupied by both struggling characters) — Its enclosing sides converge behind the two torsos and helm; used as Tight architectural context enclosing the physical clash.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural dusk ambience appropriate to the wheelhouse is held in tense low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nRender ONE photorealistic empty location photograph, 16:9, that this shot can be staged inside later. No text or watermarks.\n\nThe THIRD attached image (STRUCTURE LOOK) is the identity source of the fixed structure at this location: its faces, openings, levels, materials and signage are truth. Where it conflicts with the LOCATION PHOTOGRAPH about the structure itself, the STRUCTURE LOOK wins; the photograph still governs the surroundings, time of day and lighting.", "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S28sh9__bgfirst_bg.png", "asset_id": "f57a75ad-c482-4f13-a23e-d94676d36f71", "input_asset_ids": [ "5940ee4c-6eca-4cf4-99ca-ea7d57b53dfd", "04e8e811-2c45-4edf-8650-64f34852aa94", "f2bc4368-d3d7-477f-b65c-0f576ca7be06" ] }, "S28sh9": { "input_fingerprint": "4c85efcc57aa7d18", "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): 조타실 안, 방향타를 쥔 인우의 팔목을 수리영이 양손으로 강하게 꽉 쥐어 제어하며 두 사람의 힘이 팽팽하게 부딪히는 찰나.\n\nLOCATION (lock): Just inside the small fishing boat's wheelhouse, at the helm where the two struggle over control at dusk. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the wheelhouse beside the helm at waist height, the camera completes a forceful push to within close range of the joined arms, using a restrained Dutch tilt to sharpen the imbalance. 수리영’s two hands clamp around 인우’s wrist in the center while his other arm and both torsos pull in opposing diagonals around the helm, preserving the physical cause of the struggle.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Helm (Held by 인우 during the struggle) — Its gripping side faces 인우, with its rim angled toward the waist-height camera; used as Central mechanical anchor around which both characters exert opposing force; Wheelhouse interior (Occupied by both struggling characters) — Its enclosing sides converge behind the two torsos and helm; used as Tight architectural context enclosing the physical clash.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural dusk ambience appropriate to the wheelhouse is held in tense low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse already contains numerous circular markings, including the red circle that will shortly become visible to Su-ri-young. Her backpack and the old photograph remain with her during the struggle.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Stage the shot. The FIRST attached image (SHOT BACKGROUND) is the finished empty background of this shot — keep it EXACTLY: its camera, perspective, architecture, lighting and every fixed feature stay untouched. The SECOND attached image (LAYOUT SKETCH) tells you ONLY where the people go: each sketched person's position, screen size, pose and the gaze/motion arrows. Ignore the sketch's background lines. The CHARACTER REFERENCE photographs show the real people.\nPlace the real people into the background at exactly the sketched positions, sizes and poses, following the arrow directions. No sketch lines or arrows may remain.\n\nCreate ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 조타실 안, 방향타를 쥔 인우의 팔목을 수리영이 양손으로 강하게 꽉 쥐어 제어하며 두 사람의 힘이 팽팽하게 부딪히는 찰나.\n\nLOCATION (lock): Just inside the small fishing boat's wheelhouse, at the helm where the two struggle over control at dusk. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the wheelhouse beside the helm at waist height, the camera completes a forceful push to within close range of the joined arms, using a restrained Dutch tilt to sharpen the imbalance. 수리영’s two hands clamp around 인우’s wrist in the center while his other arm and both torsos pull in opposing diagonals around the helm, preserving the physical cause of the struggle.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Helm (Held by 인우 during the struggle) — Its gripping side faces 인우, with its rim angled toward the waist-height camera; used as Central mechanical anchor around which both characters exert opposing force; Wheelhouse interior (Occupied by both struggling characters) — Its enclosing sides converge behind the two torsos and helm; used as Tight architectural context enclosing the physical clash.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural dusk ambience appropriate to the wheelhouse is held in tense low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse already contains numerous circular markings, including the red circle that will shortly become visible to Su-ri-young. Her backpack and the old photograph remain with her during the struggle.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "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): 조타실 안, 방향타를 쥔 인우의 팔목을 수리영이 양손으로 강하게 꽉 쥐어 제어하며 두 사람의 힘이 팽팽하게 부딪히는 찰나.\n\nLOCATION (lock): Just inside the small fishing boat's wheelhouse, at the helm where the two struggle over control at dusk. 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\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Inside the wheelhouse beside the helm at waist height, the camera completes a forceful push to within close range of the joined arms, using a restrained Dutch tilt to sharpen the imbalance. 수리영’s two hands clamp around 인우’s wrist in the center while his other arm and both torsos pull in opposing diagonals around the helm, preserving the physical cause of the struggle.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: Helm (Held by 인우 during the struggle) — Its gripping side faces 인우, with its rim angled toward the waist-height camera; used as Central mechanical anchor around which both characters exert opposing force; Wheelhouse interior (Occupied by both struggling characters) — Its enclosing sides converge behind the two torsos and helm; used as Tight architectural context enclosing the physical clash.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Natural dusk ambience appropriate to the wheelhouse is held in tense low-key contrast with a restrained muted-neutral grade.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The wheelhouse already contains numerous circular markings, including the red circle that will shortly become visible to Su-ri-young. Her backpack and the old photograph remain with her during the struggle.\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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere." }, "roll_refs": { "A": [ { "label": "SHOT BACKGROUND", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S28sh9__bgfirst_bg.png" }, { "label": "LAYOUT SKETCH (people placement only)", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/conti/conti_S28sh9.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ], "B": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L17B01.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/79670d0c-2096-4d40-88cf-30b407411548/images/background_chain/seed_bg_fishing_vessel_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "CHARACTER REFERENCE — 인우: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" } ] }, "judge_flip": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "인물 레퍼런스의 의상을 충실히 반영하였고, 지시된 낡은 사진 소품과 배낭을 정확히 구현하여 프롬프트 일치도가 높습니다." }, { "label": "A", "score": 6, "verdict_ko": "인우의 의상 레퍼런스(파란 재킷)가 전혀 반영되지 않았으며, 지시된 낡은 사진 소품도 누락되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "인물 레퍼런스의 의상을 충실히 반영하였고, 지시된 낡은 사진 소품과 배낭을 정확히 구현하여 프롬프트 일치도가 높습니다." }, { "label": "A", "score": 6, "verdict_ko": "인우의 의상 레퍼런스(파란 재킷)가 전혀 반영되지 않았으며, 지시된 낡은 사진 소품도 누락되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "요청된 양손 제어 동작, 인우의 파란색 재킷 의상, 그리고 수리영의 배낭 및 오래된 사진 소품이 모두 정확하게 구현되어 프롬프트 충실도가 높습니다." }, { "label": "B", "score": 6, "verdict_ko": "수리영이 양손으로 팔목을 쥐는 동작은 구현되었으나, 인우의 지정된 파란색 재킷 의상이 누락되었고 수리영이 지녀야 할 낡은 사진 소품도 보이지 않습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "요청된 양손 제어 동작, 인우의 파란색 재킷 의상, 그리고 수리영의 배낭 및 오래된 사진 소품이 모두 정확하게 구현되어 프롬프트 충실도가 높습니다." }, { "label": "A", "score": 6, "verdict_ko": "수리영이 양손으로 팔목을 쥐는 동작은 구현되었으나, 인우의 지정된 파란색 재킷 의상이 누락되었고 수리영이 지녀야 할 낡은 사진 소품도 보이지 않습니다." } ] }, "combined": { "totals": { "A": 12, "B": 18 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 } }, "totals": { "A": 12, "B": 18 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "인물 레퍼런스의 의상을 충실히 반영하였고, 지시된 낡은 사진 소품과 배낭을 정확히 구현하여 프롬프트 일치도가 높습니다." }, { "label": "A", "score": 6, "verdict_ko": "인우의 의상 레퍼런스(파란 재킷)가 전혀 반영되지 않았으며, 지시된 낡은 사진 소품도 누락되었습니다." } ], "refs": [ { "label": "LOCATION PHOTOGRAPH — the exact place of this shot: its spatial layout, surroundings, fixed features, time of day and lighting mood are spatial truth; stage the moment inside this place. If a STRUCTURE LOOK photograph is also attached, that photo wins for the fixed structure itself — this photograph wins for everything around it. Never copy its camera framing.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/episodes/026b24c0-3e76-4953-ab00-129645f4e903/images/background_chain/L17B01.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/79670d0c-2096-4d40-88cf-30b407411548/images/background_chain/seed_bg_fishing_vessel_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 양손으로 인우의 팔목을 쥐어야 한다는 지시와 달리, 정체불명의 여분 팔(회색 소매)을 잡고 있으며 팔의 해부학적 구조가 엉켜 있습니다.", "fix_en": "Redraw Su-ri-young's arms and hands so that both of her hands directly and tightly clamp around In-woo's wrist, entirely removing the disconnected third arm.", "unfixable": false }, { "issue_ko": "수리영이 캐릭터 레퍼런스에 없는 겉옷(재킷)을 후드티 위에 덧입고 있습니다.", "fix_en": "Remove the heavy outer jacket from Su-ri-young so she is wearing only the grey hoodie from her character reference.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "issue_ko": "프레임이 두 얼굴과 상반신을 넓게 담은 미디엄 클로즈업에 가까워, 중앙의 결합된 팔과 손목이 지배해야 하는 지정 클로즈업보다 느슨하다.", "fix_en": "Crop tighter around Su-ri-young’s two hands clamping In-woo’s wrist and the helm, making the joined arms the central dominant subject while retaining portions of both opposing torsos.", "unfixable": false }, { "issue_ko": "후면 창틀과 조타실 수평선이 거의 반듯해 지정된 절제된 더치 틸트가 보이지 않는다.", "fix_en": "Apply a subtle Dutch rotation to the existing frame and trim the exposed corners, preserving the current subjects and geometry.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw Su-ri-young's arms and hands so that both of her hands directly and tightly clamp around In-woo's wrist, entirely removing the disconnected third arm.\n- Remove the heavy outer jacket from Su-ri-young so she is wearing only the grey hoodie from her character reference.\n- Crop tighter around Su-ri-young’s two hands clamping In-woo’s wrist and the helm, making the joined arms the central dominant subject while retaining portions of both opposing torsos.\n- Apply a subtle Dutch rotation to the existing frame and trim the exposed corners, preserving the current subjects and geometry.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 조타실 배경과 인물 외형, 소품을 잘 재현했으며 양손으로 팔목을 제어하는 핵심 동작을 무난하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "하드 위반: 남성 캐릭터의 얼굴에 검은색 사각형이 렌더링되었고 이미지 가장자리에 흰색 패널 여백이 있어 실격입니다." } ] }, "forward_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 조타실 배경과 인물 외형, 소품을 잘 재현했으며 양손으로 팔목을 제어하는 핵심 동작을 무난하게 구현했습니다." }, { "label": "B", "score": 3, "verdict_ko": "하드 위반: 남성 캐릭터의 얼굴에 검은색 사각형이 렌더링되었고 이미지 가장자리에 흰색 패널 여백이 있어 실격입니다." } ] }, "reverse_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 2, "verdict_ko": "남성 캐릭터의 얼굴이 검은색 상자로 가려져 있어 인물을 온전한 사람으로 묘사해야 하는 핵심 지침을 위반했습니다." }, { "label": "B", "score": 8, "verdict_ko": "요구된 카메라 구도(더치 틸트, 클로즈업)를 잘 따랐으며, 방향타를 둔 신체적 대립과 수리영의 소품(배낭, 사진)을 정확히 구현했습니다." } ] }, "reverse_normalized": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "B", "score": 2, "verdict_ko": "남성 캐릭터의 얼굴이 검은색 상자로 가려져 있어 인물을 온전한 사람으로 묘사해야 하는 핵심 지침을 위반했습니다." }, { "label": "A", "score": 8, "verdict_ko": "요구된 카메라 구도(더치 틸트, 클로즈업)를 잘 따랐으며, 방향타를 둔 신체적 대립과 수리영의 소품(배낭, 사진)을 정확히 구현했습니다." } ] }, "combined": { "totals": { "A": 15, "B": 5 }, "ranking": [ "A", "B" ], "agreement": true, "policy": 1 }, "winner": "A", "fix_won": false, "policy": "fix_rejudge_v1_flip" }, "bgfirst": { "bg_path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S28sh9__bgfirst_bg.png", "bg_asset_id": "f57a75ad-c482-4f13-a23e-d94676d36f71", "bg_record_key": "S28sh9::bgfirst_bg", "chain_winner": false, "authority": "plate", "seed_attached": true }, "ref_mode": "플레이트+엔티티 (2택1: 무콘티 승)", "share_plan": { "ref_plan": "background" }, "lane_policy": "ab_select_ready", "plate_select": { "candidates": { "A": "L17B01", "B": "L17B02" }, "assigned": "L17B01", "choice": "Candidate A", "confident": true, "reason_ko": "지문에서 '조타실 안'과 '방향타를 쥔'이라고 명시되어 있으므로, 조타실 내부와 조타핸들이 직접적으로 보이는 Candidate A가 적합합니다.", "kept": "L17B01" } }, "S29sh1::variants": { "author_fp": "f0df9bc371df526f", "author": { "variants": [ { "approach_ko": "깊은 초점과 엄격한 측면 구도로 방향타를 중심으로 맞선 두 사람의 힘과 공간적 대칭을 선명하게 강조한다.", "prompt_en": "Use a restrained, rectilinear wide-lens feel with deep focus, preserving the strict lateral tableau and clean separation of both full-body profiles around the helm. Let dim dusk light skim their faces, clothing folds, and the wheel’s worn surfaces from the side, creating crisp opposing contours against the dark wheelhouse while muted red and gold details remain subdued. Freeze the physical strain sharply so the taut arm, braced feet, backward lean, and central wheel read as one precise line of force." }, { "approach_ko": "압축된 원근감과 얕은 초점, 미세한 동작 흔들림으로 방향타 양쪽의 몸싸움을 더 즉각적이고 불안하게 포착한다.", "prompt_en": "Give the wide profile frame a subtly compressed lens feel, visually tightening the space between the opposing bodies and the wheel without changing their placement. Keep both figures and the helm legible on their shared midground plane while allowing the enclosing wheelhouse to fall into soft, low-key dusk shadow. Preserve crisp faces and planted body weight, but permit a trace of natural motion softness along the pulling hands and taut arm, making the struggle feel caught at its most unstable instant." } ] }, "reused": false }, "S29sh1": { "input_fingerprint": "b680833b9e2d4500", "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): 어둑한 조타실 내부, 방향타를 사이에 두고 인우의 팔을 팽팽하게 잡아당기고 있는 수리영의 측면 전신.\n\nLOCATION (lock): Inside the dim wheelhouse of the small fishing boat, directly beside the helm at dusk. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height beside the confrontation, hold a static wide profile composition with 수리영 on the left and 인우 on the right, their full bodies divided by the steering wheel at frame center. 수리영 leans backward as she pulls his taut arm across the wheel, while 인우 counters the force with his weight set away from her; the shared lateral axis makes the wheel read as both obstacle and point of struggle.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, reaches for 인우's arm across the steering wheel; 인우 in the middle-right of the frame, midground; steering wheel separating them in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: steering wheel (Positioned between 수리영 and 인우 during the struggle) — Its edge and side profile face the camera, with the two figures pulling across opposite sides; used as Central compositional divider and physical fulcrum for the struggle; wheelhouse interior (Darkened at dusk); used as Encloses the figures while leaving their opposing silhouettes legible.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dim dusk ambience holds the wheelhouse in low-key contrast, with muted neutrals and restrained deep red and gold accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The numerous circular markings, including the red circle, remain fixed throughout the wheelhouse. Su-ri-young still has her backpack and the old photograph as she restrains In-woo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 어둑한 조타실 내부, 방향타를 사이에 두고 인우의 팔을 팽팽하게 잡아당기고 있는 수리영의 측면 전신.\n\nLOCATION (lock): Inside the dim wheelhouse of the small fishing boat, directly beside the helm at dusk. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height beside the confrontation, hold a static wide profile composition with 수리영 on the left and 인우 on the right, their full bodies divided by the steering wheel at frame center. 수리영 leans backward as she pulls his taut arm across the wheel, while 인우 counters the force with his weight set away from her; the shared lateral axis makes the wheel read as both obstacle and point of struggle.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, reaches for 인우's arm across the steering wheel; 인우 in the middle-right of the frame, midground; steering wheel separating them in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: steering wheel (Positioned between 수리영 and 인우 during the struggle) — Its edge and side profile face the camera, with the two figures pulling across opposite sides; used as Central compositional divider and physical fulcrum for the struggle; wheelhouse interior (Darkened at dusk); used as Encloses the figures while leaving their opposing silhouettes legible.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dim dusk ambience holds the wheelhouse in low-key contrast, with muted neutrals and restrained deep red and gold accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The numerous circular markings, including the red circle, remain fixed throughout the wheelhouse. Su-ri-young still has her backpack and the old photograph as she restrains In-woo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained, rectilinear wide-lens feel with deep focus, preserving the strict lateral tableau and clean separation of both full-body profiles around the helm. Let dim dusk light skim their faces, clothing folds, and the wheel’s worn surfaces from the side, creating crisp opposing contours against the dark wheelhouse while muted red and gold details remain subdued. Freeze the physical strain sharply so the taut arm, braced feet, backward lean, and central wheel read as one precise line of force.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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): 어둑한 조타실 내부, 방향타를 사이에 두고 인우의 팔을 팽팽하게 잡아당기고 있는 수리영의 측면 전신.\n\nLOCATION (lock): Inside the dim wheelhouse of the small fishing boat, directly beside the helm at dusk. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: From waist height beside the confrontation, hold a static wide profile composition with 수리영 on the left and 인우 on the right, their full bodies divided by the steering wheel at frame center. 수리영 leans backward as she pulls his taut arm across the wheel, while 인우 counters the force with his weight set away from her; the shared lateral axis makes the wheel read as both obstacle and point of struggle.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: 수리영 in the middle-left of the frame, midground, reaches for 인우's arm across the steering wheel; 인우 in the middle-right of the frame, midground; steering wheel separating them in the middle-center of the frame, midground.\n- KEY BACKGROUND ELEMENTS: steering wheel (Positioned between 수리영 and 인우 during the struggle) — Its edge and side profile face the camera, with the two figures pulling across opposite sides; used as Central compositional divider and physical fulcrum for the struggle; wheelhouse interior (Darkened at dusk); used as Encloses the figures while leaving their opposing silhouettes legible.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dim dusk ambience holds the wheelhouse in low-key contrast, with muted neutrals and restrained deep red and gold accents.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The numerous circular markings, including the red circle, remain fixed throughout the wheelhouse. Su-ri-young still has her backpack and the old photograph as she restrains In-woo.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nGive the wide profile frame a subtly compressed lens feel, visually tightening the space between the opposing bodies and the wheel without changing their placement. Keep both figures and the helm legible on their shared midground plane while allowing the enclosing wheelhouse to fall into soft, low-key dusk shadow. Preserve crisp faces and planted body weight, but permit a trace of natural motion softness along the pulling hands and taut arm, making the struggle feel caught at its most unstable 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": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지시된 프레이밍과 측면 전신 구도를 안정적인 신체 구조로 잘 구현했으나, 배경에 있어야 할 원형 마커들이 누락되었습니다." }, { "label": "B", "score": 3, "verdict_ko": "배경의 원형 마커는 반영되었으나, 수리영의 손이 3개로 묘사되는 등 물리적으로 불가능한 신체 구조 오류(하드 위반)가 발생했습니다." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S28sh9_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": "방향타(스티어링 휠)의 측면이 카메라를 향해야 한다는 프롬프트 지시와 달리, 방향타가 정면을 향해 벽에 평면적으로 배치되어 있습니다.", "fix_en": "Rotate the steering wheel and its mounting structure 90 degrees so its edge and side profile face the camera.", "unfixable": true }, { "issue_ko": "수리영이 인우의 팔을 잡고 있는 부분에서 두 사람의 손과 손목이 해부학적으로 알아볼 수 없게 하나로 뭉개져 있습니다.", "fix_en": "Redraw the point of contact to clearly define Su-ri-young's anatomically correct fingers gripping In-woo's wrist/arm.", "unfixable": false }, { "issue_ko": "화면 좌측 상단 창틀에 붙은 종이와 방향타 위쪽 콘솔에 프롬프트에서 금지한 가짜 문자(텍스트)가 포함되어 있습니다.", "fix_en": "Erase all gibberish text and pseudo-lettering from the paper on the left wall and the console above the steering wheel.", "unfixable": false }, { "issue_ko": "레퍼런스(이전 샷)에서 지퍼가 열려 있던 인우의 파란색 재킷이 완전히 닫혀 있어, 안에 입어야 할 남색 티셔츠가 보이지 않습니다.", "fix_en": "Open the front of In-woo's blue jacket to reveal the plain navy blue t-shirt he is wearing underneath.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Extend the frame downward to include both figures’ complete feet while preserving their current scale, positions, and the existing camera geometry.", "issue_ko": "와이드 전신 구도여야 하지만 두 인물의 발이 화면 아래에서 잘려 전신이 온전히 보이지 않는다.", "unfixable": false }, { "fix_en": "Make the steering wheel read less front-on and closer to an edge-on side profile within the existing frame, without changing the camera position or surrounding geometry.", "issue_ko": "계약은 카메라를 대치 축의 측면에 두고 방향타의 가장자리와 측면이 보이게 요구하지만, 현재 사진은 방향타의 원형 정면을 곧바로 바라보는 카메라 축이다.", "unfixable": true } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw the point of contact to clearly define Su-ri-young's anatomically correct fingers gripping In-woo's wrist/arm.\n- Erase all gibberish text and pseudo-lettering from the paper on the left wall and the console above the steering wheel.\n- Open the front of In-woo's blue jacket to reveal the plain navy blue t-shirt he is wearing underneath.\n- Extend the frame downward to include both figures’ complete feet while preserving their current scale, positions, and the existing camera geometry.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 측면 전신 구도를 정확히 구현해 발끝까지 프레임에 담아냈으며, 의상 디테일과 맞잡은 손의 형태가 자연스럽습니다." }, { "label": "A", "score": 6, "verdict_ko": "요구된 전신 샷임에도 발목이 잘려 있으며, 인물들이 팔을 잡고 있는 손 부분의 형태가 다소 부자연스럽게 묘사되었습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 8, "verdict_ko": "지시된 측면 전신 구도를 정확히 구현해 발끝까지 프레임에 담아냈으며, 의상 디테일과 맞잡은 손의 형태가 자연스럽습니다." }, { "label": "A", "score": 6, "verdict_ko": "요구된 전신 샷임에도 발목이 잘려 있으며, 인물들이 팔을 잡고 있는 손 부분의 형태가 다소 부자연스럽게 묘사되었습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 7, "verdict_ko": "지정된 와이드 샷 구도와 전신 대치 상황을 정확히 구현하였고, 참조 이미지의 소품과 의상 디테일을 잘 유지했습니다." }, { "label": "B", "score": 4, "verdict_ko": "좌측 벽면에 금지된 텍스트가 생성되었고, 두 인물이 맞잡은 손의 해부학적 형태가 무너져 감점 요인이 되었습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 7, "verdict_ko": "지정된 와이드 샷 구도와 전신 대치 상황을 정확히 구현하였고, 참조 이미지의 소품과 의상 디테일을 잘 유지했습니다." }, { "label": "A", "score": 4, "verdict_ko": "좌측 벽면에 금지된 텍스트가 생성되었고, 두 인물이 맞잡은 손의 해부학적 형태가 무너져 감점 요인이 되었습니다." } ] }, "combined": { "totals": { "A": 10, "B": 15 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S28sh9" } }, "S29sh5::variants": { "author_fp": "a638c3a352cd3ed8", "author": { "variants": [ { "approach_ko": "증거를 확인하듯 절제된 관찰 시점으로, 눌리는 접점과 사진의 마모를 선명하게 강조한다.", "prompt_en": "Use a restrained, evidentiary insert with a natural close-focus lens feel and minimal perspective distortion. Hold the focus plane across the fingertip’s point of compression and the still-visible contours of the child’s face, resolving skin pressure, paper fibers, creases, and worn emulsion while the rest of the photograph and wheelhouse recede into soft dusk blur. Let the existing dim ambience rake gently across the contact point, emphasizing physical pressure without theatrical highlights." }, { "approach_ko": "주관적이고 긴장된 초근접 시선으로, 얕은 초점과 사선의 손가락이 인식의 충격을 만든다.", "prompt_en": "Make the insert feel subjective and tense through an extremely shallow, intimate focus falloff. Use the diagonal finger as a strong visual lead into the lower-center contact point, with only the compressed patch of the pictured face and the fingertip edge reaching crisp focus; allow the surrounding print and nearby controls to dissolve rapidly into muted dusk softness. Preserve dense shadows and restrained color so the identification lands as a sudden, private realization." } ] }, "reused": true }, "S29sh5": { "input_fingerprint": "7759b8ffb92530cf", "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): 수리영의 검지손가락이 낡은 사진 속 어린아이의 얼굴 부분을 꾹 내리누른 찰나의 클로즈업.\n\nLOCATION (lock): At the helm area inside the boat's dim wheelhouse, focused on the old photograph held between the two people. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Place the camera immediately above and slightly beside the photograph, isolating the pressed fingertip and the child's face while keeping the photograph within the lower third rather than letting it overwhelm the frame. 수리영's index finger enters diagonally from the upper edge and compresses the pictured face at the instant of identification, with the surrounding image falling softly out of focus.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: worn photograph with the child's face under 수리영's fingertip in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 낡은 사진 (Worn and being pressed by 수리영's index finger) — The image-bearing front face is visible to camera, with one 어린 여자아이의 face directly beneath 수리영's fingertip; used as Primary content surface beneath the pressing fingertip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key dusk ambience preserves the photograph's legibility without introducing a separate visible light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 wheelhouse lighting, nearby control materials, and both occupants' unchanged clothing from the reference. Exclude the wide steering-wheel struggle and their full bodies, framing only the finger pressing the child's face in the old photograph.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old photograph remains in Su-ri-young’s hand as she points to the child she identifies as Min-sook; the wheelhouse’s many circles remain in place around them.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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): 수리영의 검지손가락이 낡은 사진 속 어린아이의 얼굴 부분을 꾹 내리누른 찰나의 클로즈업.\n\nLOCATION (lock): At the helm area inside the boat's dim wheelhouse, focused on the old photograph held between the two people. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Place the camera immediately above and slightly beside the photograph, isolating the pressed fingertip and the child's face while keeping the photograph within the lower third rather than letting it overwhelm the frame. 수리영's index finger enters diagonally from the upper edge and compresses the pictured face at the instant of identification, with the surrounding image falling softly out of focus.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: worn photograph with the child's face under 수리영's fingertip in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 낡은 사진 (Worn and being pressed by 수리영's index finger) — The image-bearing front face is visible to camera, with one 어린 여자아이의 face directly beneath 수리영's fingertip; used as Primary content surface beneath the pressing fingertip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key dusk ambience preserves the photograph's legibility without introducing a separate visible light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 wheelhouse lighting, nearby control materials, and both occupants' unchanged clothing from the reference. Exclude the wide steering-wheel struggle and their full bodies, framing only the finger pressing the child's face in the old photograph.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old photograph remains in Su-ri-young’s hand as she points to the child she identifies as Min-sook; the wheelhouse’s many circles remain in place around them.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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, evidentiary insert with a natural close-focus lens feel and minimal perspective distortion. Hold the focus plane across the fingertip’s point of compression and the still-visible contours of the child’s face, resolving skin pressure, paper fibers, creases, and worn emulsion while the rest of the photograph and wheelhouse recede into soft dusk blur. Let the existing dim ambience rake gently across the contact point, emphasizing physical pressure without theatrical 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.", "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): 수리영의 검지손가락이 낡은 사진 속 어린아이의 얼굴 부분을 꾹 내리누른 찰나의 클로즈업.\n\nLOCATION (lock): At the helm area inside the boat's dim wheelhouse, focused on the old photograph held between the two people. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Place the camera immediately above and slightly beside the photograph, isolating the pressed fingertip and the child's face while keeping the photograph within the lower third rather than letting it overwhelm the frame. 수리영's index finger enters diagonally from the upper edge and compresses the pictured face at the instant of identification, with the surrounding image falling softly out of focus.\n- FRAMING SCALE: insert close-up on a detail\n- FRAME LAYOUT: worn photograph with the child's face under 수리영's fingertip in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 낡은 사진 (Worn and being pressed by 수리영's index finger) — The image-bearing front face is visible to camera, with one 어린 여자아이의 face directly beneath 수리영's fingertip; used as Primary content surface beneath the pressing fingertip.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Low-key dusk ambience preserves the photograph's legibility without introducing a separate visible light source.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\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 wheelhouse lighting, nearby control materials, and both occupants' unchanged clothing from the reference. Exclude the wide steering-wheel struggle and their full bodies, framing only the finger pressing the child's face in the old photograph.\n\nREALIZE FIGURATIVE LANGUAGE AS A LIVE-ACTION SHOT: the Korean shot\ntext may describe characters metaphorically, figuratively or with\nexaggeration. Photograph what a real movie camera would actually\nrecord on a physical set — exaggerated or figurative impressions\nbecome realistic staging choices (lighting, distance, angle,\nwardrobe), not literal fantasy imagery.\nEVERY CHARACTER IS A HUMAN BEING: unless the story explicitly\nfeatures non-human or virtual beings (as in science fiction or\nfantasy), every character — however indirectly, vaguely or\nfiguratively the text describes them — IS a real human. When the\ntext gives no direct visual description of a person, IMAGINE one\nand still show them as a concrete, fully-formed human being:\nalways render the human form to the maximum extent — build,\nposture, face, hands, clothing — never reduce a person to a shape,\nblob, solid silhouette or abstract mass.\n\nEXPRESSIONS ARE ACTED, NEVER ANATOMICAL: when the text describes a\nperson's eyes, face or presence emotionally or figuratively (vacant,\nhollow, dazed, burning, lifeless gaze and the like), it describes an\nACTOR'S PERFORMANCE captured by a real camera — realize it ONLY\nthrough gaze direction, focus, eyelids, facial muscles, stillness\nand posture. Every living person's eyes remain anatomically normal\nhuman eyes with a natural iris and pupil, natural sclera and normal\nproportions; NEVER whiten, blank out, cloud over, glow, enlarge or\notherwise alter eyeballs, skin or anatomy — unless the story\nexplicitly declares that being non-human or supernatural in form.\n\nPROPS FACE THE RIGHT WAY: every handheld or used object must be\noriented exactly as its real-world use requires. A person reading,\nwatching or operating something (a phone, a photograph, a paper,\nany device) has its functional side — screen, front, page — facing\nTHEIR OWN eyes; the camera then sees whatever side the staging\ngeometry implies (often its back). Show the functional side to the\ncamera ONLY when the shot text itself stages it toward the viewer.\nNever flip, mirror or reverse an object's front and back.\n\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The old photograph remains in Su-ri-young’s hand as she points to the child she identifies as Min-sook; the wheelhouse’s many circles remain in place around them.\n\nPEOPLE: the SHOT TEXT alone decides whether any person is visible in this shot. IF a person appears, they must be one of: 수리영 (한국인 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text 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):\nMake the insert feel subjective and tense through an extremely shallow, intimate focus falloff. Use the diagonal finger as a strong visual lead into the lower-center contact point, with only the compressed patch of the pictured face and the fingertip edge reaching crisp focus; allow the surrounding print and nearby controls to dissolve rapidly into muted dusk softness. Preserve dense shadows and restrained color so the identification lands as a sudden, private realization.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins." }, "totals": { "B": 9, "A": 3 }, "selected": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "제공된 사진 소품의 디테일(인물 구도, 의상, 텍스트)을 매우 정확하게 재현했으며, 요구된 인서트 클로즈업 앵글과 손가락이 얼굴을 누르는 동작을 완벽하게 구현했습니다." }, { "label": "A", "score": 3, "verdict_ko": "지시된 구도는 어느 정도 따랐으나, 레퍼런스로 제공된 사진 소품과 전혀 다른 새로운 사진(소녀의 얼굴만 크게 나온 사진)을 생성하여 소품 일치도에서 크게 실패했습니다." } ], "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/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S29sh1_sel.png" }, { "label": "CHARACTER REFERENCE — 수리영: the exact person appearing in this shot; match face, hair and build exactly.", "path": "" }, { "label": "PROP REFERENCE — 피 묻은 낡은 어린아이 사진: the exact object appearing in this shot; match its look, material and wear exactly.", "path": "" } ], "critique": { "issues": [ { "issue_ko": "검지손가락과 손등이 20대 초반 여성(수리영)의 것이 아닌, 털이 있고 주름진 거친 남성의 손으로 묘사되었습니다.", "fix_en": "Change the hand and finger to be smooth and youthful, matching a woman in her early 20s, and remove the knuckle hair.", "unfixable": false }, { "issue_ko": "사진이 수리영의 손에 들려 있어야 함에도 조종 패널 위에 평평하게 놓여 있으며, 배경의 다이얼들에 비해 사진이 비현실적으로 거대하게 왜곡되었습니다.", "fix_en": "Show the photograph being held in a hand rather than resting on the console, and shrink it to a realistic physical size for a handheld print.", "unfixable": true }, { "issue_ko": "사진 속 소녀의 얼굴과 배경 이미지가 세로가 아닌 가로 방향으로 90도 눕혀져 인쇄되어 있습니다.", "fix_en": "Rotate the image content within the photograph by 90 degrees so the child's face is upright.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Outpaint additional wheelhouse background above and around the existing composition so the photograph occupies only the lower third while its pressed face remains at lower-center.", "issue_ko": "낡은 사진이 화면의 하단 절반 이상을 차지해, 사진을 하단 1/3 안에 두고 프레임을 압도하지 않게 하라는 구도보다 지나치게 크게 보인다.", "unfixable": false } ] }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S29sh1" } }, "S29sh12::variants": { "author_fp": "fd605bdfb1392acf", "author": { "variants": [ { "approach_ko": "압축된 망원 질감과 매우 얕은 심도로 인우의 냉정한 시선과 두 인물의 짓눌린 밀착감을 강조한다.", "prompt_en": "Use a compressed, intimate lens feel with very shallow depth of field, holding the profile and forward-looking eye in crisp focus while the overlapping figure falls gently softer toward the left. Let the distant circular forms dissolve into muted, ominous bokeh, with cool dusk grazing the facial plane and a restrained red accent barely registering behind him. Keep the image still, dense, and claustrophobic." }, { "approach_ko": "근거리의 완만한 광각과 층이 살아 있는 심도로 인물의 반응과 선실 속 원형 모티프를 한 공간 안에 연결한다.", "prompt_en": "Use a close, subtly wider lens feel that preserves the fixed profile framing while giving the cramped interior stronger spatial depth. Maintain clear facial detail and believable weight in the collapsed overlap, allowing the deeper circular forms to remain softly recognizable rather than fully blurred; guide attention through layered planes from the left foreground to the stare on the right and then into the dim background. Shape the dusk illumination with restrained low-key falloff, cool neutral surfaces, and a small controlled red presence." } ] }, "reused": false }, "S29sh12": { "input_fingerprint": "c1aec8664585ab41", "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): 어깨에 축 늘어진 수리영을 그대로 둔 채 서늘한 눈빛으로 앞을 노려보는 인우의 측면 얼굴.\n\nLOCATION (lock): Inside the dim wheelhouse beside the helm, where the unconscious passenger collapses onto the boatman's shoulder. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At face height beside 인우, tighten into a static close profile so his face occupies the right half while 수리영's collapsed head and upper torso remain draped across his near shoulder on the left. Keep his cold forward stare sharp and allow the circular forms deeper in the wheelhouse to remain secondary, preserving the ominous visual connection without pulling focus from his reaction.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 수많은 원형과 붉은 원 (Visible deeper inside the wheelhouse, including the newly discovered red circle); used as Soft secondary shapes beyond 인우's profile, carrying the visual motif from the preceding push into the wheelhouse; wheelhouse interior (Darkened at dusk); used as Confined background around the two overlapping figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dim dusk ambience and low-key contrast hold 인우's expression in cool restraint while muted neutrals and a restrained red accent sustain the occult tension.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cramped wheelhouse, controls, dim sunset lighting, and both occupants' unchanged clothing from the reference. Exclude the earlier arm-pulling struggle, the prominently displayed photograph, and the woman's standing pose; she must now hang limp against the man's shoulder.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Her unconscious body hangs completely limp against Inwoo’s shoulder, with her upper body collapsed onto him and her head and limbs slack.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s unconscious body hangs completely limp against Inwoo’s shoulder, with her upper body collapsed onto him and her head and limbs slack; her backpack remains with her aboard the boat. The numerous wheelhouse circles, including the red one, remain fixed 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.", "roll_prompts": { "A": "Create ONE FINAL photorealistic live-action film still of the moment below — Incheon, South Korea, 2026; all people are Korean unless stated. TIME OF DAY (lock): dusk.\n\nSHOT TEXT (authoritative, Korean): 어깨에 축 늘어진 수리영을 그대로 둔 채 서늘한 눈빛으로 앞을 노려보는 인우의 측면 얼굴.\n\nLOCATION (lock): Inside the dim wheelhouse beside the helm, where the unconscious passenger collapses onto the boatman's shoulder. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At face height beside 인우, tighten into a static close profile so his face occupies the right half while 수리영's collapsed head and upper torso remain draped across his near shoulder on the left. Keep his cold forward stare sharp and allow the circular forms deeper in the wheelhouse to remain secondary, preserving the ominous visual connection without pulling focus from his reaction.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 수많은 원형과 붉은 원 (Visible deeper inside the wheelhouse, including the newly discovered red circle); used as Soft secondary shapes beyond 인우's profile, carrying the visual motif from the preceding push into the wheelhouse; wheelhouse interior (Darkened at dusk); used as Confined background around the two overlapping figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dim dusk ambience and low-key contrast hold 인우's expression in cool restraint while muted neutrals and a restrained red accent sustain the occult tension.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cramped wheelhouse, controls, dim sunset lighting, and both occupants' unchanged clothing from the reference. Exclude the earlier arm-pulling struggle, the prominently displayed photograph, and the woman's standing pose; she must now hang limp against the man's shoulder.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Her unconscious body hangs completely limp against Inwoo’s shoulder, with her upper body collapsed onto him and her head and limbs slack.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s unconscious body hangs completely limp against Inwoo’s shoulder, with her upper body collapsed onto him and her head and limbs slack; her backpack remains with her aboard the boat. The numerous wheelhouse circles, including the red one, remain fixed 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a compressed, intimate lens feel with very shallow depth of field, holding the profile and forward-looking eye in crisp focus while the overlapping figure falls gently softer toward the left. Let the distant circular forms dissolve into muted, ominous bokeh, with cool dusk grazing the facial plane and a restrained red accent barely registering behind him. Keep the image still, dense, and claustrophobic.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the no-text rule — is binding\nexactly as written earlier. If the VISUAL APPROACH appears to conflict\nwith any of them, the earlier instruction wins.", "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): 어깨에 축 늘어진 수리영을 그대로 둔 채 서늘한 눈빛으로 앞을 노려보는 인우의 측면 얼굴.\n\nLOCATION (lock): Inside the dim wheelhouse beside the helm, where the unconscious passenger collapses onto the boatman's shoulder. The shot takes place here — the attached PREVIOUS SHOT STILL shows this exact place.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: At face height beside 인우, tighten into a static close profile so his face occupies the right half while 수리영's collapsed head and upper torso remain draped across his near shoulder on the left. Keep his cold forward stare sharp and allow the circular forms deeper in the wheelhouse to remain secondary, preserving the ominous visual connection without pulling focus from his reaction.\n- FRAMING SCALE: close-up\n- KEY BACKGROUND ELEMENTS: 수많은 원형과 붉은 원 (Visible deeper inside the wheelhouse, including the newly discovered red circle); used as Soft secondary shapes beyond 인우's profile, carrying the visual motif from the preceding push into the wheelhouse; wheelhouse interior (Darkened at dusk); used as Confined background around the two overlapping figures.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: Dim dusk ambience and low-key contrast hold 인우's expression in cool restraint while muted neutrals and a restrained red accent sustain the occult tension.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the frame.\n\nTHIS SHOT CONTINUES THE PREVIOUS SHOT: everything the attached still established — the place, its fixed features and wear, each person's clothing and state — persists. Any person in it who cannot move stays PRECISELY as photographed (body, pose, contact points, held objects); only the camera changes.\n\nPREVIOUS STILL USAGE (follow exactly — what to take from the attached still and what to exclude): Take the same cramped wheelhouse, controls, dim sunset lighting, and both occupants' unchanged clothing from the reference. Exclude the earlier arm-pulling struggle, the prominently displayed photograph, and the woman's standing pose; she must now hang limp against the man's shoulder.\n\nIMMOBILE CHARACTER POSE — CANONICAL (identical wherever this character appears in ANY panel; on any conflict THIS POSE WINS): Her unconscious body hangs completely limp against Inwoo’s shoulder, with her upper body collapsed onto him and her head and limbs slack.\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\nNATURAL PERFORMANCE (default only — every explicit direction above\nwins): when no pose contract, immobility contract or shot-text\ndirection says otherwise, people read as alive in mid-moment —\nbelievable weight shift, hands engaged with something real, gaze on\nthe target the SHOT TEXT implies; avoid a stiff attention stance (feet\ntogether, arms hanging straight down) and a blank stare into the lens.\nIf the SHOT TEXT or any POSE/IMMOBILE contract above stages stillness,\ndeath, sleep, unconsciousness, restraint, drill or an explicit\ndirect-to-camera look, follow THAT exactly — this clause never\noverrides it.\n\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): Su-ri-young’s unconscious body hangs completely limp against Inwoo’s shoulder, with her upper body collapsed onto him and her head and limbs slack; her backpack remains with her aboard the boat. The numerous wheelhouse circles, including the red one, remain fixed 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 20대 초반 여성, 자연스러운 타원형 얼굴, 짙은 검은색 머리카락, 중간 길이 헤어스타일); 인우 (한국인 20대 중반 남성, 매끈한 얼굴선, 짙은 검은색 머리카락, 수려한 청년 얼굴) — never anyone else, and never add a person the shot text does not show. IF only part of a person is in frame (a hand, arm, foot, back, silhouette), that body part belongs to the specific person the shot text names — its sex, age, build, skin and grooming must unmistakably match that person's profile above.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a close, subtly wider lens feel that preserves the fixed profile framing while giving the cramped interior stronger spatial depth. Maintain clear facial detail and believable weight in the collapsed overlap, allowing the deeper circular forms to remain softly recognizable rather than fully blurred; guide attention through layered planes from the left foreground to the stare on the right and then into the dim background. Shape the dusk illumination with restrained low-key falloff, cool neutral surfaces, and a small controlled red 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": 5, "B": 3 }, "selected": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 5, "verdict_ko": "인우의 측면 클로즈업과 조명 분위기는 훌륭하게 구현했으나, 수리영이 어깨가 아닌 조종반에 엎드려 있어 핵심 자세 지시를 완전히 놓침." }, { "label": "B", "score": 3, "verdict_ko": "수리영의 머리와 팔이 인우의 어깨와 겹치는 위치가 물리적으로 불가능한 기형적 인체 구조를 낳아 심각한 규정 위반(Hard Violation)에 해당함." } ], "refs": [ { "label": "PREVIOUS SHOT STILL — a visually related earlier shot of this same place: the location's look, materials, fixed features, lighting mood and each person's clothing are LOCKED to this photo; never copy its camera framing. If this photo shows a character who cannot move (dead or unconscious), that character's exact pose, position and orientation are ALSO LOCKED — treat their body as a fixed prop of the set that only the camera moves around.", "path": "/Users/manta/Documents/Projects/TheRoad-I1/projects/79670d0c-2096-4d40-88cf-30b407411548/images/026b24c0-3e76-4953-ab00-129645f4e903/scene/recipe/S29sh5_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 인우의 어깨에 기대지 않고 조종판 위에 엎드려 있습니다.", "fix_en": "Redraw Su-ri-young's head and upper torso so that her unconscious body hangs completely limp against Inwoo's left shoulder, removing her from the control panel.", "unfixable": false } ] }, "composition_critique": { "issues": [] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Redraw Su-ri-young's head and upper torso so that her unconscious body hangs completely limp against Inwoo's left shoulder, removing her from the control panel.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "수리영이 인우의 어깨가 아닌 조타실 콘솔 위에 엎드려 있어 '어깨에 축 늘어진'이라는 핵심적인 자세 지침을 위반했습니다." }, { "label": "B", "score": 9, "verdict_ko": "수리영이 인우의 어깨에 기댄 채 의식을 잃은 모습과 인우의 서늘한 측면 응시를 프레임 내에 정확하게 구현했으며, 배경의 붉은 원과 조명 분위기도 잘 살렸습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "A", "score": 3, "verdict_ko": "수리영이 인우의 어깨가 아닌 조타실 콘솔 위에 엎드려 있어 '어깨에 축 늘어진'이라는 핵심적인 자세 지침을 위반했습니다." }, { "label": "B", "score": 9, "verdict_ko": "수리영이 인우의 어깨에 기댄 채 의식을 잃은 모습과 인우의 서늘한 측면 응시를 프레임 내에 정확하게 구현했으며, 배경의 붉은 원과 조명 분위기도 잘 살렸습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "인우의 측면 얼굴과 그의 어깨에 기대어 축 늘어진 수리영의 자세, 배경의 붉은 원 등 프롬프트의 요구사항을 매우 충실히 구현했습니다." }, { "label": "B", "score": 4, "verdict_ko": "수리영이 인우의 어깨가 아닌 조종석 패널 위에 엎드려 있어, 필수적인 자세(어깨에 늘어짐) 지침을 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "인우의 측면 얼굴과 그의 어깨에 기대어 축 늘어진 수리영의 자세, 배경의 붉은 원 등 프롬프트의 요구사항을 매우 충실히 구현했습니다." }, { "label": "A", "score": 4, "verdict_ko": "수리영이 인우의 어깨가 아닌 조종석 패널 위에 엎드려 있어, 필수적인 자세(어깨에 늘어짐) 지침을 위반했습니다." } ] }, "combined": { "totals": { "A": 7, "B": 18 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "prev+엔티티", "share_plan": { "ref_plan": "prev", "prev_anchor_tag": "S29sh5" } }, "S30sh3::variants": { "author_fp": "7aa8c4e8c29a1a7a", "author": { "variants": [ { "approach_ko": "망원 압축감으로 어선과 안개 장벽의 거리를 눌러, 배가 거대한 안개에 삼켜지는 위압감을 강조한다.", "prompt_en": "Use a restrained telephoto feel to compress the boat against the immense fog boundary, making the fog read as a heavy, nearly solid wall while preserving the locked wide composition. Keep the sea visually flattened and subdued, with the stern crisp against the softer obscured bow; let the low sunset reflection form a muted golden path through the otherwise cool neutral field. Render the wake as a long, clean directional scar leading straight into the vessel and the ominous fog." }, { "approach_ko": "넓고 깊은 원근감으로 전경의 긴 항적에서 안개 속 뱃머리까지 시선을 빠르게 끌어가는 추적 숏으로 본다.", "prompt_en": "Use a broad, deep-perspective lens feel that gives the foreground wake strong visual reach, pulling the eye from the bottom edge along the boat’s exact forward axis and into the fog. Maintain clear depth separation between the textured foreground water, the small midground stern, and the horizon-erasing fog bank, with natural atmospheric falloff progressively softening the bow. Allow the sunset’s grounded gold to shimmer selectively on the disturbed water while the fog remains dense, restrained, and ominously diffuse." } ] }, "reused": false }, "S30sh3": { "input_fingerprint": "0a9deb15598261e8", "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, foggy.\n\nSHOT TEXT (authoritative, Korean): 뱃머리가 짙고 거대한 안개 띠의 경계선 속으로 반쯤 가려진 채, 선미 뒤로 하얀 항적을 길게 남기며 물살을 가르고 있는 소형 어선의 뒷모습.\n\nLOCATION (lock): On the open sea at dusk, at the boundary of a vast dense fog bank swallowing the small fishing boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track directly behind the fishing boat from a moderately elevated position, keeping the stern in the lower center, the long white wake receding toward the bottom edge, and the bow advancing into the fog boundary above. The boat remains under a third of the frame so the erased horizon and massive fog band carry equal narrative weight, while its forward axis runs from lower center toward upper center.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: fishing boat and its forward travel axis in the middle-center of the frame, midground, moves toward dense fog bank; dense fog bank boundary in the upper-center of the frame, background; long white wake in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 소형 어선 (Moving forward into the fog) — Its stern faces the camera while its bow points away and is partly concealed by the fog bank; used as Tracked primary vessel and scale reference against the fog bank; 하얀 항적 (Extending behind the stern across the sea); used as Linear foreground trace of the boat's movement; 짙고 거대한 안개 띠 (Dense, with the fishing boat entering its boundary); used as Upper-frame boundary that progressively conceals the bow and obscures the horizon; 바다 (Being cut by the moving fishing boat) — The sea reflects the setting sun as a golden shadow around the boat's path; used as Broad environmental field beneath the boat and carrier of the sunset reflection; 지는 해 (Low and gradually setting); used as Distant contextual element establishing the late-day setting.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The setting sun casts a grounded golden reflection across the sea while the dense fog and restrained neutral palette deepen the boundary's ominous weight.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The unconscious Su-ri-young remains aboard In-woo’s boat as it enters the fog, with her backpack and printed maps still with her. The wheelhouse’s circular markings remain on the vessel.\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, foggy.\n\nSHOT TEXT (authoritative, Korean): 뱃머리가 짙고 거대한 안개 띠의 경계선 속으로 반쯤 가려진 채, 선미 뒤로 하얀 항적을 길게 남기며 물살을 가르고 있는 소형 어선의 뒷모습.\n\nLOCATION (lock): On the open sea at dusk, at the boundary of a vast dense fog bank swallowing the small fishing boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track directly behind the fishing boat from a moderately elevated position, keeping the stern in the lower center, the long white wake receding toward the bottom edge, and the bow advancing into the fog boundary above. The boat remains under a third of the frame so the erased horizon and massive fog band carry equal narrative weight, while its forward axis runs from lower center toward upper center.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: fishing boat and its forward travel axis in the middle-center of the frame, midground, moves toward dense fog bank; dense fog bank boundary in the upper-center of the frame, background; long white wake in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 소형 어선 (Moving forward into the fog) — Its stern faces the camera while its bow points away and is partly concealed by the fog bank; used as Tracked primary vessel and scale reference against the fog bank; 하얀 항적 (Extending behind the stern across the sea); used as Linear foreground trace of the boat's movement; 짙고 거대한 안개 띠 (Dense, with the fishing boat entering its boundary); used as Upper-frame boundary that progressively conceals the bow and obscures the horizon; 바다 (Being cut by the moving fishing boat) — The sea reflects the setting sun as a golden shadow around the boat's path; used as Broad environmental field beneath the boat and carrier of the sunset reflection; 지는 해 (Low and gradually setting); used as Distant contextual element establishing the late-day setting.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The setting sun casts a grounded golden reflection across the sea while the dense fog and restrained neutral palette deepen the boundary's ominous weight.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The unconscious Su-ri-young remains aboard In-woo’s boat as it enters the fog, with her backpack and printed maps still with her. The wheelhouse’s circular markings remain on the vessel.\n\nNO PEOPLE IN THIS SHOT: the shot text shows only the place and its state — no living person or any body part appears in frame, unless the shot text itself explicitly says so.\n\nNo text, captions, watermarks or annotations anywhere.\n\nVISUAL APPROACH — this take (camera, framing, composition):\nUse a restrained telephoto feel to compress the boat against the immense fog boundary, making the fog read as a heavy, nearly solid wall while preserving the locked wide composition. Keep the sea visually flattened and subdued, with the stern crisp against the softer obscured bow; let the low sunset reflection form a muted golden path through the otherwise cool neutral field. Render the wake as a long, clean directional scar leading straight into the vessel and the ominous fog.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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, foggy.\n\nSHOT TEXT (authoritative, Korean): 뱃머리가 짙고 거대한 안개 띠의 경계선 속으로 반쯤 가려진 채, 선미 뒤로 하얀 항적을 길게 남기며 물살을 가르고 있는 소형 어선의 뒷모습.\n\nLOCATION (lock): On the open sea at dusk, at the boundary of a vast dense fog bank swallowing the small fishing boat. The shot takes place here — the attached LOCATION PHOTOGRAPH shows the exact spot.\n\nCAMERA & FRAME (follow exactly — this is the composition authority for this still; the reference images supply identity and place, never the framing):\n- CAMERA: Track directly behind the fishing boat from a moderately elevated position, keeping the stern in the lower center, the long white wake receding toward the bottom edge, and the bow advancing into the fog boundary above. The boat remains under a third of the frame so the erased horizon and massive fog band carry equal narrative weight, while its forward axis runs from lower center toward upper center.\n- FRAMING SCALE: wide shot\n- FRAME LAYOUT: fishing boat and its forward travel axis in the middle-center of the frame, midground, moves toward dense fog bank; dense fog bank boundary in the upper-center of the frame, background; long white wake in the lower-center of the frame, foreground.\n- KEY BACKGROUND ELEMENTS: 소형 어선 (Moving forward into the fog) — Its stern faces the camera while its bow points away and is partly concealed by the fog bank; used as Tracked primary vessel and scale reference against the fog bank; 하얀 항적 (Extending behind the stern across the sea); used as Linear foreground trace of the boat's movement; 짙고 거대한 안개 띠 (Dense, with the fishing boat entering its boundary); used as Upper-frame boundary that progressively conceals the bow and obscures the horizon; 바다 (Being cut by the moving fishing boat) — The sea reflects the setting sun as a golden shadow around the boat's path; used as Broad environmental field beneath the boat and carrier of the sunset reflection; 지는 해 (Low and gradually setting); used as Distant contextual element establishing the late-day setting.\nCompose the frame exactly as specified above — camera angle, subject scale and screen placement. Keep true physical scale between people and background elements: fixtures and distant objects occupy only the small screen area their real size and distance dictate; never enlarge a background object into a foreground presence.\n\nLIGHTING & MOOD (these govern light, color and surface rendering):\n- LIGHTING & MOOD: The setting sun casts a grounded golden reflection across the sea while the dense fog and restrained neutral palette deepen the boundary's ominous weight.\n- MATERIAL REALISM: every object and surface must read as a real,\n physical material — correct texture, weight, wear and light response\n (metal reflects, fabric drapes and creases, liquid is glossy and\n pools, painted or drawn marks sit ON a surface and follow its\n curvature and lighting). Nothing may look like a flat sticker, a\n doodle or a graphic overlay pasted onto the 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\nDRAWN MARKS KEEP THEIR SHAPE: any mark the text describes as drawn,\npainted or traced (a circle, a line, a symbol) is a STROKE sitting on\nthe surface — an outline whose interior still shows the underlying\nsurface (wall, skin, paper). Render its stated shape faithfully: a\ndrawn circle stays an open ring of brush-width, never filled into a\nsolid disc, unless the text explicitly says it is filled.\n\nCARRIED STATE (persist exactly — must match the neighbouring shots of this scene): The unconscious Su-ri-young remains aboard In-woo’s boat as it enters the fog, with her backpack and printed maps still with her. The wheelhouse’s circular markings remain on the vessel.\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 broad, deep-perspective lens feel that gives the foreground wake strong visual reach, pulling the eye from the bottom edge along the boat’s exact forward axis and into the fog. Maintain clear depth separation between the textured foreground water, the small midground stern, and the horizon-erasing fog bank, with natural atmospheric falloff progressively softening the bow. Allow the sunset’s grounded gold to shimmer selectively on the disturbed water while the fog remains dense, restrained, and ominously diffuse.\n\nPRIORITY OF INSTRUCTIONS: the VISUAL APPROACH above governs camera,\nframing and composition ONLY. Every other instruction in this prompt —\nlocation, time of day, people and their identity, poses, props,\ncarried state, physical rules, and the 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": "프레이밍과 구도가 지시사항과 잘 일치하며, 특히 어선의 뱃머리가 안개 띠 속으로 반쯤 가려진 모습을 효과적으로 연출했습니다." }, { "label": "A", "score": 4, "verdict_ko": "카메라 구도와 항적의 표현은 좋으나, 배가 아직 안개에 진입하지 않아 뱃머리가 가려진다는 핵심 묘사를 충족하지 못했습니다." } ], "refs": [], "critique": { "issues": [ { "issue_ko": "프롬프트는 뱃머리가 안개 띠 속으로 반쯤 가려져야 한다고 명시했으나, 어선 전체가 안개 밖에 선명하게 노출되어 있습니다.", "fix_en": "Add dense fog overlapping the front half of the fishing boat so that the bow is half-hidden and swallowed by the fog bank.", "unfixable": false }, { "issue_ko": "조타실 뒷면의 원형 표시 안에 'Y'라는 알파벳이 있고 지붕 쪽에도 문자가 생성되어 '텍스트 불가' 규칙을 위반했습니다.", "fix_en": "Remove the letter 'Y' from inside the circular marking on the wheelhouse and the small letter on the structure above it, leaving only plain painted circles.", "unfixable": false } ] }, "composition_critique": { "issues": [ { "fix_en": "Extend the fog boundary forward over the boat’s upper/front section so the bow area is progressively concealed while preserving the stern and wake.", "issue_ko": "어선이 안개 경계에 진입해 뱃머리가 부분적으로 가려져야 하지만, 현재는 배 전체 윤곽이 안개 앞에서 선명하게 드러나 있다.", "unfixable": false }, { "fix_en": "Densify the fog across the distant waterline to fully erase the visible horizon while preserving the existing boat, wake, and viewpoint.", "issue_ko": "안개 띠가 수평선을 지워야 하지만 화면 양측과 배 뒤쪽에 먼 수평선이 아직 뚜렷하게 보인다.", "unfixable": false } ] }, "fix_prompt": "Edit this photograph. Apply ONLY the corrections listed below,\neach in place, changing nothing else:\n\nCORRECTIONS:\n- Add dense fog overlapping the front half of the fishing boat so that the bow is half-hidden and swallowed by the fog bank.\n- Remove the letter 'Y' from inside the circular marking on the wheelhouse and the small letter on the structure above it, leaving only plain painted circles.\n- Extend the fog boundary forward over the boat’s upper/front section so the bow area is progressively concealed while preserving the stern and wake.\n- Densify the fog across the distant waterline to fully erase the visible horizon while preserving the existing boat, wake, and viewpoint.\n\nPRESERVE EVERYTHING ELSE EXACTLY as in the original photograph —\nevery person, pose, face, garment, object, set feature, lighting,\ncamera and framing stay identical. No text, captions or\nwatermarks anywhere.", "fix_rejudge": { "forward_raw": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "텍스트 금지 규칙을 준수하며 카메라 구도, 짙은 안개 속으로 들어가는 어선의 뒷모습, 하얀 항적 및 시간대 등 모든 지시 사항을 완벽하게 구현했습니다." }, { "label": "A", "score": 8, "verdict_ko": "전반적인 구도와 분위기는 매우 훌륭하나, 조타실의 원형 표식 내부에 알파벳(텍스트)이 생성되어 '텍스트 생성 금지' 규칙을 위반했습니다." } ] }, "forward_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 10, "verdict_ko": "텍스트 금지 규칙을 준수하며 카메라 구도, 짙은 안개 속으로 들어가는 어선의 뒷모습, 하얀 항적 및 시간대 등 모든 지시 사항을 완벽하게 구현했습니다." }, { "label": "A", "score": 8, "verdict_ko": "전반적인 구도와 분위기는 매우 훌륭하나, 조타실의 원형 표식 내부에 알파벳(텍스트)이 생성되어 '텍스트 생성 금지' 규칙을 위반했습니다." } ] }, "reverse_raw": { "winner": "A", "ranking": [ "A", "B" ], "verdicts": [ { "label": "A", "score": 9, "verdict_ko": "텍스트 생성 금지 규칙과 속이 빈 원형 마크에 대한 지시를 정확히 따랐으며, 지정된 후면 구도와 안개 속으로 향하는 어선의 모습을 사실적으로 구현했습니다." }, { "label": "B", "score": 6, "verdict_ko": "구도와 분위기는 훌륭하게 구현했으나, 조타실의 원형 마크와 상단에 알파벳 문자가 생성되어 '텍스트 생성 금지' 지시를 위반했습니다." } ] }, "reverse_normalized": { "winner": "B", "ranking": [ "B", "A" ], "verdicts": [ { "label": "B", "score": 9, "verdict_ko": "텍스트 생성 금지 규칙과 속이 빈 원형 마크에 대한 지시를 정확히 따랐으며, 지정된 후면 구도와 안개 속으로 향하는 어선의 모습을 사실적으로 구현했습니다." }, { "label": "A", "score": 6, "verdict_ko": "구도와 분위기는 훌륭하게 구현했으나, 조타실의 원형 마크와 상단에 알파벳 문자가 생성되어 '텍스트 생성 금지' 지시를 위반했습니다." } ] }, "combined": { "totals": { "A": 14, "B": 19 }, "ranking": [ "B", "A" ], "agreement": true, "policy": 1 }, "winner": "B", "fix_won": true, "policy": "fix_rejudge_v1_flip" }, "ref_mode": "플레이트만 (배경 전용)", "share_plan": { "ref_plan": "background" } } }